1 /* 2 * linux/fs/namei.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 */ 6 7 /* 8 * Some corrections by tytso. 9 */ 10 11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 12 * lookup logic. 13 */ 14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 15 */ 16 17 #include <linux/init.h> 18 #include <linux/export.h> 19 #include <linux/kernel.h> 20 #include <linux/slab.h> 21 #include <linux/fs.h> 22 #include <linux/namei.h> 23 #include <linux/pagemap.h> 24 #include <linux/fsnotify.h> 25 #include <linux/personality.h> 26 #include <linux/security.h> 27 #include <linux/ima.h> 28 #include <linux/syscalls.h> 29 #include <linux/mount.h> 30 #include <linux/audit.h> 31 #include <linux/capability.h> 32 #include <linux/file.h> 33 #include <linux/fcntl.h> 34 #include <linux/device_cgroup.h> 35 #include <linux/fs_struct.h> 36 #include <linux/posix_acl.h> 37 #include <linux/hash.h> 38 #include <asm/uaccess.h> 39 40 #include "internal.h" 41 #include "mount.h" 42 43 /* [Feb-1997 T. Schoebel-Theuer] 44 * Fundamental changes in the pathname lookup mechanisms (namei) 45 * were necessary because of omirr. The reason is that omirr needs 46 * to know the _real_ pathname, not the user-supplied one, in case 47 * of symlinks (and also when transname replacements occur). 48 * 49 * The new code replaces the old recursive symlink resolution with 50 * an iterative one (in case of non-nested symlink chains). It does 51 * this with calls to <fs>_follow_link(). 52 * As a side effect, dir_namei(), _namei() and follow_link() are now 53 * replaced with a single function lookup_dentry() that can handle all 54 * the special cases of the former code. 55 * 56 * With the new dcache, the pathname is stored at each inode, at least as 57 * long as the refcount of the inode is positive. As a side effect, the 58 * size of the dcache depends on the inode cache and thus is dynamic. 59 * 60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 61 * resolution to correspond with current state of the code. 62 * 63 * Note that the symlink resolution is not *completely* iterative. 64 * There is still a significant amount of tail- and mid- recursion in 65 * the algorithm. Also, note that <fs>_readlink() is not used in 66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 67 * may return different results than <fs>_follow_link(). Many virtual 68 * filesystems (including /proc) exhibit this behavior. 69 */ 70 71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 73 * and the name already exists in form of a symlink, try to create the new 74 * name indicated by the symlink. The old code always complained that the 75 * name already exists, due to not following the symlink even if its target 76 * is nonexistent. The new semantics affects also mknod() and link() when 77 * the name is a symlink pointing to a non-existent name. 78 * 79 * I don't know which semantics is the right one, since I have no access 80 * to standards. But I found by trial that HP-UX 9.0 has the full "new" 81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 82 * "old" one. Personally, I think the new semantics is much more logical. 83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing 84 * file does succeed in both HP-UX and SunOs, but not in Solaris 85 * and in the old Linux semantics. 86 */ 87 88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 89 * semantics. See the comments in "open_namei" and "do_link" below. 90 * 91 * [10-Sep-98 Alan Modra] Another symlink change. 92 */ 93 94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 95 * inside the path - always follow. 96 * in the last component in creation/removal/renaming - never follow. 97 * if LOOKUP_FOLLOW passed - follow. 98 * if the pathname has trailing slashes - follow. 99 * otherwise - don't follow. 100 * (applied in that order). 101 * 102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 104 * During the 2.4 we need to fix the userland stuff depending on it - 105 * hopefully we will be able to get rid of that wart in 2.5. So far only 106 * XEmacs seems to be relying on it... 107 */ 108 /* 109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 111 * any extra contention... 112 */ 113 114 /* In order to reduce some races, while at the same time doing additional 115 * checking and hopefully speeding things up, we copy filenames to the 116 * kernel data space before using them.. 117 * 118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 119 * PATH_MAX includes the nul terminator --RR. 120 */ 121 122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 123 124 struct filename * 125 getname_flags(const char __user *filename, int flags, int *empty) 126 { 127 struct filename *result; 128 char *kname; 129 int len; 130 131 result = audit_reusename(filename); 132 if (result) 133 return result; 134 135 result = __getname(); 136 if (unlikely(!result)) 137 return ERR_PTR(-ENOMEM); 138 139 /* 140 * First, try to embed the struct filename inside the names_cache 141 * allocation 142 */ 143 kname = (char *)result->iname; 144 result->name = kname; 145 146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 147 if (unlikely(len < 0)) { 148 __putname(result); 149 return ERR_PTR(len); 150 } 151 152 /* 153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 154 * separate struct filename so we can dedicate the entire 155 * names_cache allocation for the pathname, and re-do the copy from 156 * userland. 157 */ 158 if (unlikely(len == EMBEDDED_NAME_MAX)) { 159 const size_t size = offsetof(struct filename, iname[1]); 160 kname = (char *)result; 161 162 /* 163 * size is chosen that way we to guarantee that 164 * result->iname[0] is within the same object and that 165 * kname can't be equal to result->iname, no matter what. 166 */ 167 result = kzalloc(size, GFP_KERNEL); 168 if (unlikely(!result)) { 169 __putname(kname); 170 return ERR_PTR(-ENOMEM); 171 } 172 result->name = kname; 173 len = strncpy_from_user(kname, filename, PATH_MAX); 174 if (unlikely(len < 0)) { 175 __putname(kname); 176 kfree(result); 177 return ERR_PTR(len); 178 } 179 if (unlikely(len == PATH_MAX)) { 180 __putname(kname); 181 kfree(result); 182 return ERR_PTR(-ENAMETOOLONG); 183 } 184 } 185 186 result->refcnt = 1; 187 /* The empty path is special. */ 188 if (unlikely(!len)) { 189 if (empty) 190 *empty = 1; 191 if (!(flags & LOOKUP_EMPTY)) { 192 putname(result); 193 return ERR_PTR(-ENOENT); 194 } 195 } 196 197 result->uptr = filename; 198 result->aname = NULL; 199 audit_getname(result); 200 return result; 201 } 202 203 struct filename * 204 getname(const char __user * filename) 205 { 206 return getname_flags(filename, 0, NULL); 207 } 208 209 struct filename * 210 getname_kernel(const char * filename) 211 { 212 struct filename *result; 213 int len = strlen(filename) + 1; 214 215 result = __getname(); 216 if (unlikely(!result)) 217 return ERR_PTR(-ENOMEM); 218 219 if (len <= EMBEDDED_NAME_MAX) { 220 result->name = (char *)result->iname; 221 } else if (len <= PATH_MAX) { 222 struct filename *tmp; 223 224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 225 if (unlikely(!tmp)) { 226 __putname(result); 227 return ERR_PTR(-ENOMEM); 228 } 229 tmp->name = (char *)result; 230 result = tmp; 231 } else { 232 __putname(result); 233 return ERR_PTR(-ENAMETOOLONG); 234 } 235 memcpy((char *)result->name, filename, len); 236 result->uptr = NULL; 237 result->aname = NULL; 238 result->refcnt = 1; 239 audit_getname(result); 240 241 return result; 242 } 243 244 void putname(struct filename *name) 245 { 246 BUG_ON(name->refcnt <= 0); 247 248 if (--name->refcnt > 0) 249 return; 250 251 if (name->name != name->iname) { 252 __putname(name->name); 253 kfree(name); 254 } else 255 __putname(name); 256 } 257 258 static int check_acl(struct inode *inode, int mask) 259 { 260 #ifdef CONFIG_FS_POSIX_ACL 261 struct posix_acl *acl; 262 263 if (mask & MAY_NOT_BLOCK) { 264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 265 if (!acl) 266 return -EAGAIN; 267 /* no ->get_acl() calls in RCU mode... */ 268 if (acl == ACL_NOT_CACHED) 269 return -ECHILD; 270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 271 } 272 273 acl = get_acl(inode, ACL_TYPE_ACCESS); 274 if (IS_ERR(acl)) 275 return PTR_ERR(acl); 276 if (acl) { 277 int error = posix_acl_permission(inode, acl, mask); 278 posix_acl_release(acl); 279 return error; 280 } 281 #endif 282 283 return -EAGAIN; 284 } 285 286 /* 287 * This does the basic permission checking 288 */ 289 static int acl_permission_check(struct inode *inode, int mask) 290 { 291 unsigned int mode = inode->i_mode; 292 293 if (likely(uid_eq(current_fsuid(), inode->i_uid))) 294 mode >>= 6; 295 else { 296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 297 int error = check_acl(inode, mask); 298 if (error != -EAGAIN) 299 return error; 300 } 301 302 if (in_group_p(inode->i_gid)) 303 mode >>= 3; 304 } 305 306 /* 307 * If the DACs are ok we don't need any capability check. 308 */ 309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 310 return 0; 311 return -EACCES; 312 } 313 314 /** 315 * generic_permission - check for access rights on a Posix-like filesystem 316 * @inode: inode to check access rights for 317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 318 * 319 * Used to check for read/write/execute permissions on a file. 320 * We use "fsuid" for this, letting us set arbitrary permissions 321 * for filesystem access without changing the "normal" uids which 322 * are used for other things. 323 * 324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 325 * request cannot be satisfied (eg. requires blocking or too much complexity). 326 * It would then be called again in ref-walk mode. 327 */ 328 int generic_permission(struct inode *inode, int mask) 329 { 330 int ret; 331 332 /* 333 * Do the basic permission checks. 334 */ 335 ret = acl_permission_check(inode, mask); 336 if (ret != -EACCES) 337 return ret; 338 339 if (S_ISDIR(inode->i_mode)) { 340 /* DACs are overridable for directories */ 341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 342 return 0; 343 if (!(mask & MAY_WRITE)) 344 if (capable_wrt_inode_uidgid(inode, 345 CAP_DAC_READ_SEARCH)) 346 return 0; 347 return -EACCES; 348 } 349 /* 350 * Read/write DACs are always overridable. 351 * Executable DACs are overridable when there is 352 * at least one exec bit set. 353 */ 354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 356 return 0; 357 358 /* 359 * Searching includes executable on directories, else just read. 360 */ 361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 362 if (mask == MAY_READ) 363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 364 return 0; 365 366 return -EACCES; 367 } 368 EXPORT_SYMBOL(generic_permission); 369 370 /* 371 * We _really_ want to just do "generic_permission()" without 372 * even looking at the inode->i_op values. So we keep a cache 373 * flag in inode->i_opflags, that says "this has not special 374 * permission function, use the fast case". 375 */ 376 static inline int do_inode_permission(struct inode *inode, int mask) 377 { 378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 379 if (likely(inode->i_op->permission)) 380 return inode->i_op->permission(inode, mask); 381 382 /* This gets set once for the inode lifetime */ 383 spin_lock(&inode->i_lock); 384 inode->i_opflags |= IOP_FASTPERM; 385 spin_unlock(&inode->i_lock); 386 } 387 return generic_permission(inode, mask); 388 } 389 390 /** 391 * __inode_permission - Check for access rights to a given inode 392 * @inode: Inode to check permission on 393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 394 * 395 * Check for read/write/execute permissions on an inode. 396 * 397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 398 * 399 * This does not check for a read-only file system. You probably want 400 * inode_permission(). 401 */ 402 int __inode_permission(struct inode *inode, int mask) 403 { 404 int retval; 405 406 if (unlikely(mask & MAY_WRITE)) { 407 /* 408 * Nobody gets write access to an immutable file. 409 */ 410 if (IS_IMMUTABLE(inode)) 411 return -EACCES; 412 } 413 414 retval = do_inode_permission(inode, mask); 415 if (retval) 416 return retval; 417 418 retval = devcgroup_inode_permission(inode, mask); 419 if (retval) 420 return retval; 421 422 return security_inode_permission(inode, mask); 423 } 424 EXPORT_SYMBOL(__inode_permission); 425 426 /** 427 * sb_permission - Check superblock-level permissions 428 * @sb: Superblock of inode to check permission on 429 * @inode: Inode to check permission on 430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 431 * 432 * Separate out file-system wide checks from inode-specific permission checks. 433 */ 434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 435 { 436 if (unlikely(mask & MAY_WRITE)) { 437 umode_t mode = inode->i_mode; 438 439 /* Nobody gets write access to a read-only fs. */ 440 if ((sb->s_flags & MS_RDONLY) && 441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 442 return -EROFS; 443 } 444 return 0; 445 } 446 447 /** 448 * inode_permission - Check for access rights to a given inode 449 * @inode: Inode to check permission on 450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 451 * 452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for 453 * this, letting us set arbitrary permissions for filesystem access without 454 * changing the "normal" UIDs which are used for other things. 455 * 456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 457 */ 458 int inode_permission(struct inode *inode, int mask) 459 { 460 int retval; 461 462 retval = sb_permission(inode->i_sb, inode, mask); 463 if (retval) 464 return retval; 465 return __inode_permission(inode, mask); 466 } 467 EXPORT_SYMBOL(inode_permission); 468 469 /** 470 * path_get - get a reference to a path 471 * @path: path to get the reference to 472 * 473 * Given a path increment the reference count to the dentry and the vfsmount. 474 */ 475 void path_get(const struct path *path) 476 { 477 mntget(path->mnt); 478 dget(path->dentry); 479 } 480 EXPORT_SYMBOL(path_get); 481 482 /** 483 * path_put - put a reference to a path 484 * @path: path to put the reference to 485 * 486 * Given a path decrement the reference count to the dentry and the vfsmount. 487 */ 488 void path_put(const struct path *path) 489 { 490 dput(path->dentry); 491 mntput(path->mnt); 492 } 493 EXPORT_SYMBOL(path_put); 494 495 struct nameidata { 496 struct path path; 497 struct qstr last; 498 struct path root; 499 struct inode *inode; /* path.dentry.d_inode */ 500 unsigned int flags; 501 unsigned seq, m_seq; 502 int last_type; 503 unsigned depth; 504 struct file *base; 505 char *saved_names[MAX_NESTED_LINKS + 1]; 506 }; 507 508 /* 509 * Path walking has 2 modes, rcu-walk and ref-walk (see 510 * Documentation/filesystems/path-lookup.txt). In situations when we can't 511 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 512 * normal reference counts on dentries and vfsmounts to transition to rcu-walk 513 * mode. Refcounts are grabbed at the last known good point before rcu-walk 514 * got stuck, so ref-walk may continue from there. If this is not successful 515 * (eg. a seqcount has changed), then failure is returned and it's up to caller 516 * to restart the path walk from the beginning in ref-walk mode. 517 */ 518 519 /** 520 * unlazy_walk - try to switch to ref-walk mode. 521 * @nd: nameidata pathwalk data 522 * @dentry: child of nd->path.dentry or NULL 523 * Returns: 0 on success, -ECHILD on failure 524 * 525 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 526 * for ref-walk mode. @dentry must be a path found by a do_lookup call on 527 * @nd or NULL. Must be called from rcu-walk context. 528 */ 529 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 530 { 531 struct fs_struct *fs = current->fs; 532 struct dentry *parent = nd->path.dentry; 533 534 BUG_ON(!(nd->flags & LOOKUP_RCU)); 535 536 /* 537 * After legitimizing the bastards, terminate_walk() 538 * will do the right thing for non-RCU mode, and all our 539 * subsequent exit cases should rcu_read_unlock() 540 * before returning. Do vfsmount first; if dentry 541 * can't be legitimized, just set nd->path.dentry to NULL 542 * and rely on dput(NULL) being a no-op. 543 */ 544 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 545 return -ECHILD; 546 nd->flags &= ~LOOKUP_RCU; 547 548 if (!lockref_get_not_dead(&parent->d_lockref)) { 549 nd->path.dentry = NULL; 550 goto out; 551 } 552 553 /* 554 * For a negative lookup, the lookup sequence point is the parents 555 * sequence point, and it only needs to revalidate the parent dentry. 556 * 557 * For a positive lookup, we need to move both the parent and the 558 * dentry from the RCU domain to be properly refcounted. And the 559 * sequence number in the dentry validates *both* dentry counters, 560 * since we checked the sequence number of the parent after we got 561 * the child sequence number. So we know the parent must still 562 * be valid if the child sequence number is still valid. 563 */ 564 if (!dentry) { 565 if (read_seqcount_retry(&parent->d_seq, nd->seq)) 566 goto out; 567 BUG_ON(nd->inode != parent->d_inode); 568 } else { 569 if (!lockref_get_not_dead(&dentry->d_lockref)) 570 goto out; 571 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 572 goto drop_dentry; 573 } 574 575 /* 576 * Sequence counts matched. Now make sure that the root is 577 * still valid and get it if required. 578 */ 579 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 580 spin_lock(&fs->lock); 581 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 582 goto unlock_and_drop_dentry; 583 path_get(&nd->root); 584 spin_unlock(&fs->lock); 585 } 586 587 rcu_read_unlock(); 588 return 0; 589 590 unlock_and_drop_dentry: 591 spin_unlock(&fs->lock); 592 drop_dentry: 593 rcu_read_unlock(); 594 dput(dentry); 595 goto drop_root_mnt; 596 out: 597 rcu_read_unlock(); 598 drop_root_mnt: 599 if (!(nd->flags & LOOKUP_ROOT)) 600 nd->root.mnt = NULL; 601 return -ECHILD; 602 } 603 604 static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 605 { 606 return dentry->d_op->d_revalidate(dentry, flags); 607 } 608 609 /** 610 * complete_walk - successful completion of path walk 611 * @nd: pointer nameidata 612 * 613 * If we had been in RCU mode, drop out of it and legitimize nd->path. 614 * Revalidate the final result, unless we'd already done that during 615 * the path walk or the filesystem doesn't ask for it. Return 0 on 616 * success, -error on failure. In case of failure caller does not 617 * need to drop nd->path. 618 */ 619 static int complete_walk(struct nameidata *nd) 620 { 621 struct dentry *dentry = nd->path.dentry; 622 int status; 623 624 if (nd->flags & LOOKUP_RCU) { 625 nd->flags &= ~LOOKUP_RCU; 626 if (!(nd->flags & LOOKUP_ROOT)) 627 nd->root.mnt = NULL; 628 629 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 630 rcu_read_unlock(); 631 return -ECHILD; 632 } 633 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 634 rcu_read_unlock(); 635 mntput(nd->path.mnt); 636 return -ECHILD; 637 } 638 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 639 rcu_read_unlock(); 640 dput(dentry); 641 mntput(nd->path.mnt); 642 return -ECHILD; 643 } 644 rcu_read_unlock(); 645 } 646 647 if (likely(!(nd->flags & LOOKUP_JUMPED))) 648 return 0; 649 650 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 651 return 0; 652 653 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 654 if (status > 0) 655 return 0; 656 657 if (!status) 658 status = -ESTALE; 659 660 path_put(&nd->path); 661 return status; 662 } 663 664 static __always_inline void set_root(struct nameidata *nd) 665 { 666 get_fs_root(current->fs, &nd->root); 667 } 668 669 static int link_path_walk(const char *, struct nameidata *); 670 671 static __always_inline unsigned set_root_rcu(struct nameidata *nd) 672 { 673 struct fs_struct *fs = current->fs; 674 unsigned seq, res; 675 676 do { 677 seq = read_seqcount_begin(&fs->seq); 678 nd->root = fs->root; 679 res = __read_seqcount_begin(&nd->root.dentry->d_seq); 680 } while (read_seqcount_retry(&fs->seq, seq)); 681 return res; 682 } 683 684 static void path_put_conditional(struct path *path, struct nameidata *nd) 685 { 686 dput(path->dentry); 687 if (path->mnt != nd->path.mnt) 688 mntput(path->mnt); 689 } 690 691 static inline void path_to_nameidata(const struct path *path, 692 struct nameidata *nd) 693 { 694 if (!(nd->flags & LOOKUP_RCU)) { 695 dput(nd->path.dentry); 696 if (nd->path.mnt != path->mnt) 697 mntput(nd->path.mnt); 698 } 699 nd->path.mnt = path->mnt; 700 nd->path.dentry = path->dentry; 701 } 702 703 /* 704 * Helper to directly jump to a known parsed path from ->follow_link, 705 * caller must have taken a reference to path beforehand. 706 */ 707 void nd_jump_link(struct nameidata *nd, struct path *path) 708 { 709 path_put(&nd->path); 710 711 nd->path = *path; 712 nd->inode = nd->path.dentry->d_inode; 713 nd->flags |= LOOKUP_JUMPED; 714 } 715 716 void nd_set_link(struct nameidata *nd, char *path) 717 { 718 nd->saved_names[nd->depth] = path; 719 } 720 EXPORT_SYMBOL(nd_set_link); 721 722 char *nd_get_link(struct nameidata *nd) 723 { 724 return nd->saved_names[nd->depth]; 725 } 726 EXPORT_SYMBOL(nd_get_link); 727 728 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 729 { 730 struct inode *inode = link->dentry->d_inode; 731 if (inode->i_op->put_link) 732 inode->i_op->put_link(link->dentry, nd, cookie); 733 path_put(link); 734 } 735 736 int sysctl_protected_symlinks __read_mostly = 0; 737 int sysctl_protected_hardlinks __read_mostly = 0; 738 739 /** 740 * may_follow_link - Check symlink following for unsafe situations 741 * @link: The path of the symlink 742 * @nd: nameidata pathwalk data 743 * 744 * In the case of the sysctl_protected_symlinks sysctl being enabled, 745 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 746 * in a sticky world-writable directory. This is to protect privileged 747 * processes from failing races against path names that may change out 748 * from under them by way of other users creating malicious symlinks. 749 * It will permit symlinks to be followed only when outside a sticky 750 * world-writable directory, or when the uid of the symlink and follower 751 * match, or when the directory owner matches the symlink's owner. 752 * 753 * Returns 0 if following the symlink is allowed, -ve on error. 754 */ 755 static inline int may_follow_link(struct path *link, struct nameidata *nd) 756 { 757 const struct inode *inode; 758 const struct inode *parent; 759 760 if (!sysctl_protected_symlinks) 761 return 0; 762 763 /* Allowed if owner and follower match. */ 764 inode = link->dentry->d_inode; 765 if (uid_eq(current_cred()->fsuid, inode->i_uid)) 766 return 0; 767 768 /* Allowed if parent directory not sticky and world-writable. */ 769 parent = nd->path.dentry->d_inode; 770 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 771 return 0; 772 773 /* Allowed if parent directory and link owner match. */ 774 if (uid_eq(parent->i_uid, inode->i_uid)) 775 return 0; 776 777 audit_log_link_denied("follow_link", link); 778 path_put_conditional(link, nd); 779 path_put(&nd->path); 780 return -EACCES; 781 } 782 783 /** 784 * safe_hardlink_source - Check for safe hardlink conditions 785 * @inode: the source inode to hardlink from 786 * 787 * Return false if at least one of the following conditions: 788 * - inode is not a regular file 789 * - inode is setuid 790 * - inode is setgid and group-exec 791 * - access failure for read and write 792 * 793 * Otherwise returns true. 794 */ 795 static bool safe_hardlink_source(struct inode *inode) 796 { 797 umode_t mode = inode->i_mode; 798 799 /* Special files should not get pinned to the filesystem. */ 800 if (!S_ISREG(mode)) 801 return false; 802 803 /* Setuid files should not get pinned to the filesystem. */ 804 if (mode & S_ISUID) 805 return false; 806 807 /* Executable setgid files should not get pinned to the filesystem. */ 808 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 809 return false; 810 811 /* Hardlinking to unreadable or unwritable sources is dangerous. */ 812 if (inode_permission(inode, MAY_READ | MAY_WRITE)) 813 return false; 814 815 return true; 816 } 817 818 /** 819 * may_linkat - Check permissions for creating a hardlink 820 * @link: the source to hardlink from 821 * 822 * Block hardlink when all of: 823 * - sysctl_protected_hardlinks enabled 824 * - fsuid does not match inode 825 * - hardlink source is unsafe (see safe_hardlink_source() above) 826 * - not CAP_FOWNER 827 * 828 * Returns 0 if successful, -ve on error. 829 */ 830 static int may_linkat(struct path *link) 831 { 832 const struct cred *cred; 833 struct inode *inode; 834 835 if (!sysctl_protected_hardlinks) 836 return 0; 837 838 cred = current_cred(); 839 inode = link->dentry->d_inode; 840 841 /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 842 * otherwise, it must be a safe source. 843 */ 844 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 845 capable(CAP_FOWNER)) 846 return 0; 847 848 audit_log_link_denied("linkat", link); 849 return -EPERM; 850 } 851 852 static __always_inline int 853 follow_link(struct path *link, struct nameidata *nd, void **p) 854 { 855 struct dentry *dentry = link->dentry; 856 int error; 857 char *s; 858 859 BUG_ON(nd->flags & LOOKUP_RCU); 860 861 if (link->mnt == nd->path.mnt) 862 mntget(link->mnt); 863 864 error = -ELOOP; 865 if (unlikely(current->total_link_count >= 40)) 866 goto out_put_nd_path; 867 868 cond_resched(); 869 current->total_link_count++; 870 871 touch_atime(link); 872 nd_set_link(nd, NULL); 873 874 error = security_inode_follow_link(link->dentry, nd); 875 if (error) 876 goto out_put_nd_path; 877 878 nd->last_type = LAST_BIND; 879 *p = dentry->d_inode->i_op->follow_link(dentry, nd); 880 error = PTR_ERR(*p); 881 if (IS_ERR(*p)) 882 goto out_put_nd_path; 883 884 error = 0; 885 s = nd_get_link(nd); 886 if (s) { 887 if (unlikely(IS_ERR(s))) { 888 path_put(&nd->path); 889 put_link(nd, link, *p); 890 return PTR_ERR(s); 891 } 892 if (*s == '/') { 893 if (!nd->root.mnt) 894 set_root(nd); 895 path_put(&nd->path); 896 nd->path = nd->root; 897 path_get(&nd->root); 898 nd->flags |= LOOKUP_JUMPED; 899 } 900 nd->inode = nd->path.dentry->d_inode; 901 error = link_path_walk(s, nd); 902 if (unlikely(error)) 903 put_link(nd, link, *p); 904 } 905 906 return error; 907 908 out_put_nd_path: 909 *p = NULL; 910 path_put(&nd->path); 911 path_put(link); 912 return error; 913 } 914 915 static int follow_up_rcu(struct path *path) 916 { 917 struct mount *mnt = real_mount(path->mnt); 918 struct mount *parent; 919 struct dentry *mountpoint; 920 921 parent = mnt->mnt_parent; 922 if (&parent->mnt == path->mnt) 923 return 0; 924 mountpoint = mnt->mnt_mountpoint; 925 path->dentry = mountpoint; 926 path->mnt = &parent->mnt; 927 return 1; 928 } 929 930 /* 931 * follow_up - Find the mountpoint of path's vfsmount 932 * 933 * Given a path, find the mountpoint of its source file system. 934 * Replace @path with the path of the mountpoint in the parent mount. 935 * Up is towards /. 936 * 937 * Return 1 if we went up a level and 0 if we were already at the 938 * root. 939 */ 940 int follow_up(struct path *path) 941 { 942 struct mount *mnt = real_mount(path->mnt); 943 struct mount *parent; 944 struct dentry *mountpoint; 945 946 read_seqlock_excl(&mount_lock); 947 parent = mnt->mnt_parent; 948 if (parent == mnt) { 949 read_sequnlock_excl(&mount_lock); 950 return 0; 951 } 952 mntget(&parent->mnt); 953 mountpoint = dget(mnt->mnt_mountpoint); 954 read_sequnlock_excl(&mount_lock); 955 dput(path->dentry); 956 path->dentry = mountpoint; 957 mntput(path->mnt); 958 path->mnt = &parent->mnt; 959 return 1; 960 } 961 EXPORT_SYMBOL(follow_up); 962 963 /* 964 * Perform an automount 965 * - return -EISDIR to tell follow_managed() to stop and return the path we 966 * were called with. 967 */ 968 static int follow_automount(struct path *path, unsigned flags, 969 bool *need_mntput) 970 { 971 struct vfsmount *mnt; 972 int err; 973 974 if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 975 return -EREMOTE; 976 977 /* We don't want to mount if someone's just doing a stat - 978 * unless they're stat'ing a directory and appended a '/' to 979 * the name. 980 * 981 * We do, however, want to mount if someone wants to open or 982 * create a file of any type under the mountpoint, wants to 983 * traverse through the mountpoint or wants to open the 984 * mounted directory. Also, autofs may mark negative dentries 985 * as being automount points. These will need the attentions 986 * of the daemon to instantiate them before they can be used. 987 */ 988 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 989 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 990 path->dentry->d_inode) 991 return -EISDIR; 992 993 current->total_link_count++; 994 if (current->total_link_count >= 40) 995 return -ELOOP; 996 997 mnt = path->dentry->d_op->d_automount(path); 998 if (IS_ERR(mnt)) { 999 /* 1000 * The filesystem is allowed to return -EISDIR here to indicate 1001 * it doesn't want to automount. For instance, autofs would do 1002 * this so that its userspace daemon can mount on this dentry. 1003 * 1004 * However, we can only permit this if it's a terminal point in 1005 * the path being looked up; if it wasn't then the remainder of 1006 * the path is inaccessible and we should say so. 1007 */ 1008 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 1009 return -EREMOTE; 1010 return PTR_ERR(mnt); 1011 } 1012 1013 if (!mnt) /* mount collision */ 1014 return 0; 1015 1016 if (!*need_mntput) { 1017 /* lock_mount() may release path->mnt on error */ 1018 mntget(path->mnt); 1019 *need_mntput = true; 1020 } 1021 err = finish_automount(mnt, path); 1022 1023 switch (err) { 1024 case -EBUSY: 1025 /* Someone else made a mount here whilst we were busy */ 1026 return 0; 1027 case 0: 1028 path_put(path); 1029 path->mnt = mnt; 1030 path->dentry = dget(mnt->mnt_root); 1031 return 0; 1032 default: 1033 return err; 1034 } 1035 1036 } 1037 1038 /* 1039 * Handle a dentry that is managed in some way. 1040 * - Flagged for transit management (autofs) 1041 * - Flagged as mountpoint 1042 * - Flagged as automount point 1043 * 1044 * This may only be called in refwalk mode. 1045 * 1046 * Serialization is taken care of in namespace.c 1047 */ 1048 static int follow_managed(struct path *path, unsigned flags) 1049 { 1050 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1051 unsigned managed; 1052 bool need_mntput = false; 1053 int ret = 0; 1054 1055 /* Given that we're not holding a lock here, we retain the value in a 1056 * local variable for each dentry as we look at it so that we don't see 1057 * the components of that value change under us */ 1058 while (managed = ACCESS_ONCE(path->dentry->d_flags), 1059 managed &= DCACHE_MANAGED_DENTRY, 1060 unlikely(managed != 0)) { 1061 /* Allow the filesystem to manage the transit without i_mutex 1062 * being held. */ 1063 if (managed & DCACHE_MANAGE_TRANSIT) { 1064 BUG_ON(!path->dentry->d_op); 1065 BUG_ON(!path->dentry->d_op->d_manage); 1066 ret = path->dentry->d_op->d_manage(path->dentry, false); 1067 if (ret < 0) 1068 break; 1069 } 1070 1071 /* Transit to a mounted filesystem. */ 1072 if (managed & DCACHE_MOUNTED) { 1073 struct vfsmount *mounted = lookup_mnt(path); 1074 if (mounted) { 1075 dput(path->dentry); 1076 if (need_mntput) 1077 mntput(path->mnt); 1078 path->mnt = mounted; 1079 path->dentry = dget(mounted->mnt_root); 1080 need_mntput = true; 1081 continue; 1082 } 1083 1084 /* Something is mounted on this dentry in another 1085 * namespace and/or whatever was mounted there in this 1086 * namespace got unmounted before lookup_mnt() could 1087 * get it */ 1088 } 1089 1090 /* Handle an automount point */ 1091 if (managed & DCACHE_NEED_AUTOMOUNT) { 1092 ret = follow_automount(path, flags, &need_mntput); 1093 if (ret < 0) 1094 break; 1095 continue; 1096 } 1097 1098 /* We didn't change the current path point */ 1099 break; 1100 } 1101 1102 if (need_mntput && path->mnt == mnt) 1103 mntput(path->mnt); 1104 if (ret == -EISDIR) 1105 ret = 0; 1106 return ret < 0 ? ret : need_mntput; 1107 } 1108 1109 int follow_down_one(struct path *path) 1110 { 1111 struct vfsmount *mounted; 1112 1113 mounted = lookup_mnt(path); 1114 if (mounted) { 1115 dput(path->dentry); 1116 mntput(path->mnt); 1117 path->mnt = mounted; 1118 path->dentry = dget(mounted->mnt_root); 1119 return 1; 1120 } 1121 return 0; 1122 } 1123 EXPORT_SYMBOL(follow_down_one); 1124 1125 static inline int managed_dentry_rcu(struct dentry *dentry) 1126 { 1127 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1128 dentry->d_op->d_manage(dentry, true) : 0; 1129 } 1130 1131 /* 1132 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1133 * we meet a managed dentry that would need blocking. 1134 */ 1135 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1136 struct inode **inode) 1137 { 1138 for (;;) { 1139 struct mount *mounted; 1140 /* 1141 * Don't forget we might have a non-mountpoint managed dentry 1142 * that wants to block transit. 1143 */ 1144 switch (managed_dentry_rcu(path->dentry)) { 1145 case -ECHILD: 1146 default: 1147 return false; 1148 case -EISDIR: 1149 return true; 1150 case 0: 1151 break; 1152 } 1153 1154 if (!d_mountpoint(path->dentry)) 1155 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1156 1157 mounted = __lookup_mnt(path->mnt, path->dentry); 1158 if (!mounted) 1159 break; 1160 path->mnt = &mounted->mnt; 1161 path->dentry = mounted->mnt.mnt_root; 1162 nd->flags |= LOOKUP_JUMPED; 1163 nd->seq = read_seqcount_begin(&path->dentry->d_seq); 1164 /* 1165 * Update the inode too. We don't need to re-check the 1166 * dentry sequence number here after this d_inode read, 1167 * because a mount-point is always pinned. 1168 */ 1169 *inode = path->dentry->d_inode; 1170 } 1171 return !read_seqretry(&mount_lock, nd->m_seq) && 1172 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1173 } 1174 1175 static int follow_dotdot_rcu(struct nameidata *nd) 1176 { 1177 struct inode *inode = nd->inode; 1178 if (!nd->root.mnt) 1179 set_root_rcu(nd); 1180 1181 while (1) { 1182 if (nd->path.dentry == nd->root.dentry && 1183 nd->path.mnt == nd->root.mnt) { 1184 break; 1185 } 1186 if (nd->path.dentry != nd->path.mnt->mnt_root) { 1187 struct dentry *old = nd->path.dentry; 1188 struct dentry *parent = old->d_parent; 1189 unsigned seq; 1190 1191 inode = parent->d_inode; 1192 seq = read_seqcount_begin(&parent->d_seq); 1193 if (read_seqcount_retry(&old->d_seq, nd->seq)) 1194 goto failed; 1195 nd->path.dentry = parent; 1196 nd->seq = seq; 1197 break; 1198 } 1199 if (!follow_up_rcu(&nd->path)) 1200 break; 1201 inode = nd->path.dentry->d_inode; 1202 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1203 } 1204 while (d_mountpoint(nd->path.dentry)) { 1205 struct mount *mounted; 1206 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1207 if (!mounted) 1208 break; 1209 nd->path.mnt = &mounted->mnt; 1210 nd->path.dentry = mounted->mnt.mnt_root; 1211 inode = nd->path.dentry->d_inode; 1212 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1213 if (read_seqretry(&mount_lock, nd->m_seq)) 1214 goto failed; 1215 } 1216 nd->inode = inode; 1217 return 0; 1218 1219 failed: 1220 nd->flags &= ~LOOKUP_RCU; 1221 if (!(nd->flags & LOOKUP_ROOT)) 1222 nd->root.mnt = NULL; 1223 rcu_read_unlock(); 1224 return -ECHILD; 1225 } 1226 1227 /* 1228 * Follow down to the covering mount currently visible to userspace. At each 1229 * point, the filesystem owning that dentry may be queried as to whether the 1230 * caller is permitted to proceed or not. 1231 */ 1232 int follow_down(struct path *path) 1233 { 1234 unsigned managed; 1235 int ret; 1236 1237 while (managed = ACCESS_ONCE(path->dentry->d_flags), 1238 unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1239 /* Allow the filesystem to manage the transit without i_mutex 1240 * being held. 1241 * 1242 * We indicate to the filesystem if someone is trying to mount 1243 * something here. This gives autofs the chance to deny anyone 1244 * other than its daemon the right to mount on its 1245 * superstructure. 1246 * 1247 * The filesystem may sleep at this point. 1248 */ 1249 if (managed & DCACHE_MANAGE_TRANSIT) { 1250 BUG_ON(!path->dentry->d_op); 1251 BUG_ON(!path->dentry->d_op->d_manage); 1252 ret = path->dentry->d_op->d_manage( 1253 path->dentry, false); 1254 if (ret < 0) 1255 return ret == -EISDIR ? 0 : ret; 1256 } 1257 1258 /* Transit to a mounted filesystem. */ 1259 if (managed & DCACHE_MOUNTED) { 1260 struct vfsmount *mounted = lookup_mnt(path); 1261 if (!mounted) 1262 break; 1263 dput(path->dentry); 1264 mntput(path->mnt); 1265 path->mnt = mounted; 1266 path->dentry = dget(mounted->mnt_root); 1267 continue; 1268 } 1269 1270 /* Don't handle automount points here */ 1271 break; 1272 } 1273 return 0; 1274 } 1275 EXPORT_SYMBOL(follow_down); 1276 1277 /* 1278 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 1279 */ 1280 static void follow_mount(struct path *path) 1281 { 1282 while (d_mountpoint(path->dentry)) { 1283 struct vfsmount *mounted = lookup_mnt(path); 1284 if (!mounted) 1285 break; 1286 dput(path->dentry); 1287 mntput(path->mnt); 1288 path->mnt = mounted; 1289 path->dentry = dget(mounted->mnt_root); 1290 } 1291 } 1292 1293 static void follow_dotdot(struct nameidata *nd) 1294 { 1295 if (!nd->root.mnt) 1296 set_root(nd); 1297 1298 while(1) { 1299 struct dentry *old = nd->path.dentry; 1300 1301 if (nd->path.dentry == nd->root.dentry && 1302 nd->path.mnt == nd->root.mnt) { 1303 break; 1304 } 1305 if (nd->path.dentry != nd->path.mnt->mnt_root) { 1306 /* rare case of legitimate dget_parent()... */ 1307 nd->path.dentry = dget_parent(nd->path.dentry); 1308 dput(old); 1309 break; 1310 } 1311 if (!follow_up(&nd->path)) 1312 break; 1313 } 1314 follow_mount(&nd->path); 1315 nd->inode = nd->path.dentry->d_inode; 1316 } 1317 1318 /* 1319 * This looks up the name in dcache, possibly revalidates the old dentry and 1320 * allocates a new one if not found or not valid. In the need_lookup argument 1321 * returns whether i_op->lookup is necessary. 1322 * 1323 * dir->d_inode->i_mutex must be held 1324 */ 1325 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1326 unsigned int flags, bool *need_lookup) 1327 { 1328 struct dentry *dentry; 1329 int error; 1330 1331 *need_lookup = false; 1332 dentry = d_lookup(dir, name); 1333 if (dentry) { 1334 if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1335 error = d_revalidate(dentry, flags); 1336 if (unlikely(error <= 0)) { 1337 if (error < 0) { 1338 dput(dentry); 1339 return ERR_PTR(error); 1340 } else { 1341 d_invalidate(dentry); 1342 dput(dentry); 1343 dentry = NULL; 1344 } 1345 } 1346 } 1347 } 1348 1349 if (!dentry) { 1350 dentry = d_alloc(dir, name); 1351 if (unlikely(!dentry)) 1352 return ERR_PTR(-ENOMEM); 1353 1354 *need_lookup = true; 1355 } 1356 return dentry; 1357 } 1358 1359 /* 1360 * Call i_op->lookup on the dentry. The dentry must be negative and 1361 * unhashed. 1362 * 1363 * dir->d_inode->i_mutex must be held 1364 */ 1365 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 1366 unsigned int flags) 1367 { 1368 struct dentry *old; 1369 1370 /* Don't create child dentry for a dead directory. */ 1371 if (unlikely(IS_DEADDIR(dir))) { 1372 dput(dentry); 1373 return ERR_PTR(-ENOENT); 1374 } 1375 1376 old = dir->i_op->lookup(dir, dentry, flags); 1377 if (unlikely(old)) { 1378 dput(dentry); 1379 dentry = old; 1380 } 1381 return dentry; 1382 } 1383 1384 static struct dentry *__lookup_hash(struct qstr *name, 1385 struct dentry *base, unsigned int flags) 1386 { 1387 bool need_lookup; 1388 struct dentry *dentry; 1389 1390 dentry = lookup_dcache(name, base, flags, &need_lookup); 1391 if (!need_lookup) 1392 return dentry; 1393 1394 return lookup_real(base->d_inode, dentry, flags); 1395 } 1396 1397 /* 1398 * It's more convoluted than I'd like it to be, but... it's still fairly 1399 * small and for now I'd prefer to have fast path as straight as possible. 1400 * It _is_ time-critical. 1401 */ 1402 static int lookup_fast(struct nameidata *nd, 1403 struct path *path, struct inode **inode) 1404 { 1405 struct vfsmount *mnt = nd->path.mnt; 1406 struct dentry *dentry, *parent = nd->path.dentry; 1407 int need_reval = 1; 1408 int status = 1; 1409 int err; 1410 1411 /* 1412 * Rename seqlock is not required here because in the off chance 1413 * of a false negative due to a concurrent rename, we're going to 1414 * do the non-racy lookup, below. 1415 */ 1416 if (nd->flags & LOOKUP_RCU) { 1417 unsigned seq; 1418 dentry = __d_lookup_rcu(parent, &nd->last, &seq); 1419 if (!dentry) 1420 goto unlazy; 1421 1422 /* 1423 * This sequence count validates that the inode matches 1424 * the dentry name information from lookup. 1425 */ 1426 *inode = dentry->d_inode; 1427 if (read_seqcount_retry(&dentry->d_seq, seq)) 1428 return -ECHILD; 1429 1430 /* 1431 * This sequence count validates that the parent had no 1432 * changes while we did the lookup of the dentry above. 1433 * 1434 * The memory barrier in read_seqcount_begin of child is 1435 * enough, we can use __read_seqcount_retry here. 1436 */ 1437 if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 1438 return -ECHILD; 1439 nd->seq = seq; 1440 1441 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 1442 status = d_revalidate(dentry, nd->flags); 1443 if (unlikely(status <= 0)) { 1444 if (status != -ECHILD) 1445 need_reval = 0; 1446 goto unlazy; 1447 } 1448 } 1449 path->mnt = mnt; 1450 path->dentry = dentry; 1451 if (likely(__follow_mount_rcu(nd, path, inode))) 1452 return 0; 1453 unlazy: 1454 if (unlazy_walk(nd, dentry)) 1455 return -ECHILD; 1456 } else { 1457 dentry = __d_lookup(parent, &nd->last); 1458 } 1459 1460 if (unlikely(!dentry)) 1461 goto need_lookup; 1462 1463 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 1464 status = d_revalidate(dentry, nd->flags); 1465 if (unlikely(status <= 0)) { 1466 if (status < 0) { 1467 dput(dentry); 1468 return status; 1469 } 1470 d_invalidate(dentry); 1471 dput(dentry); 1472 goto need_lookup; 1473 } 1474 1475 path->mnt = mnt; 1476 path->dentry = dentry; 1477 err = follow_managed(path, nd->flags); 1478 if (unlikely(err < 0)) { 1479 path_put_conditional(path, nd); 1480 return err; 1481 } 1482 if (err) 1483 nd->flags |= LOOKUP_JUMPED; 1484 *inode = path->dentry->d_inode; 1485 return 0; 1486 1487 need_lookup: 1488 return 1; 1489 } 1490 1491 /* Fast lookup failed, do it the slow way */ 1492 static int lookup_slow(struct nameidata *nd, struct path *path) 1493 { 1494 struct dentry *dentry, *parent; 1495 int err; 1496 1497 parent = nd->path.dentry; 1498 BUG_ON(nd->inode != parent->d_inode); 1499 1500 mutex_lock(&parent->d_inode->i_mutex); 1501 dentry = __lookup_hash(&nd->last, parent, nd->flags); 1502 mutex_unlock(&parent->d_inode->i_mutex); 1503 if (IS_ERR(dentry)) 1504 return PTR_ERR(dentry); 1505 path->mnt = nd->path.mnt; 1506 path->dentry = dentry; 1507 err = follow_managed(path, nd->flags); 1508 if (unlikely(err < 0)) { 1509 path_put_conditional(path, nd); 1510 return err; 1511 } 1512 if (err) 1513 nd->flags |= LOOKUP_JUMPED; 1514 return 0; 1515 } 1516 1517 static inline int may_lookup(struct nameidata *nd) 1518 { 1519 if (nd->flags & LOOKUP_RCU) { 1520 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1521 if (err != -ECHILD) 1522 return err; 1523 if (unlazy_walk(nd, NULL)) 1524 return -ECHILD; 1525 } 1526 return inode_permission(nd->inode, MAY_EXEC); 1527 } 1528 1529 static inline int handle_dots(struct nameidata *nd, int type) 1530 { 1531 if (type == LAST_DOTDOT) { 1532 if (nd->flags & LOOKUP_RCU) { 1533 if (follow_dotdot_rcu(nd)) 1534 return -ECHILD; 1535 } else 1536 follow_dotdot(nd); 1537 } 1538 return 0; 1539 } 1540 1541 static void terminate_walk(struct nameidata *nd) 1542 { 1543 if (!(nd->flags & LOOKUP_RCU)) { 1544 path_put(&nd->path); 1545 } else { 1546 nd->flags &= ~LOOKUP_RCU; 1547 if (!(nd->flags & LOOKUP_ROOT)) 1548 nd->root.mnt = NULL; 1549 rcu_read_unlock(); 1550 } 1551 } 1552 1553 /* 1554 * Do we need to follow links? We _really_ want to be able 1555 * to do this check without having to look at inode->i_op, 1556 * so we keep a cache of "no, this doesn't need follow_link" 1557 * for the common case. 1558 */ 1559 static inline int should_follow_link(struct dentry *dentry, int follow) 1560 { 1561 return unlikely(d_is_symlink(dentry)) ? follow : 0; 1562 } 1563 1564 static inline int walk_component(struct nameidata *nd, struct path *path, 1565 int follow) 1566 { 1567 struct inode *inode; 1568 int err; 1569 /* 1570 * "." and ".." are special - ".." especially so because it has 1571 * to be able to know about the current root directory and 1572 * parent relationships. 1573 */ 1574 if (unlikely(nd->last_type != LAST_NORM)) 1575 return handle_dots(nd, nd->last_type); 1576 err = lookup_fast(nd, path, &inode); 1577 if (unlikely(err)) { 1578 if (err < 0) 1579 goto out_err; 1580 1581 err = lookup_slow(nd, path); 1582 if (err < 0) 1583 goto out_err; 1584 1585 inode = path->dentry->d_inode; 1586 } 1587 err = -ENOENT; 1588 if (d_is_negative(path->dentry)) 1589 goto out_path_put; 1590 1591 if (should_follow_link(path->dentry, follow)) { 1592 if (nd->flags & LOOKUP_RCU) { 1593 if (unlikely(unlazy_walk(nd, path->dentry))) { 1594 err = -ECHILD; 1595 goto out_err; 1596 } 1597 } 1598 BUG_ON(inode != path->dentry->d_inode); 1599 return 1; 1600 } 1601 path_to_nameidata(path, nd); 1602 nd->inode = inode; 1603 return 0; 1604 1605 out_path_put: 1606 path_to_nameidata(path, nd); 1607 out_err: 1608 terminate_walk(nd); 1609 return err; 1610 } 1611 1612 /* 1613 * This limits recursive symlink follows to 8, while 1614 * limiting consecutive symlinks to 40. 1615 * 1616 * Without that kind of total limit, nasty chains of consecutive 1617 * symlinks can cause almost arbitrarily long lookups. 1618 */ 1619 static inline int nested_symlink(struct path *path, struct nameidata *nd) 1620 { 1621 int res; 1622 1623 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1624 path_put_conditional(path, nd); 1625 path_put(&nd->path); 1626 return -ELOOP; 1627 } 1628 BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1629 1630 nd->depth++; 1631 current->link_count++; 1632 1633 do { 1634 struct path link = *path; 1635 void *cookie; 1636 1637 res = follow_link(&link, nd, &cookie); 1638 if (res) 1639 break; 1640 res = walk_component(nd, path, LOOKUP_FOLLOW); 1641 put_link(nd, &link, cookie); 1642 } while (res > 0); 1643 1644 current->link_count--; 1645 nd->depth--; 1646 return res; 1647 } 1648 1649 /* 1650 * We can do the critical dentry name comparison and hashing 1651 * operations one word at a time, but we are limited to: 1652 * 1653 * - Architectures with fast unaligned word accesses. We could 1654 * do a "get_unaligned()" if this helps and is sufficiently 1655 * fast. 1656 * 1657 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1658 * do not trap on the (extremely unlikely) case of a page 1659 * crossing operation. 1660 * 1661 * - Furthermore, we need an efficient 64-bit compile for the 1662 * 64-bit case in order to generate the "number of bytes in 1663 * the final mask". Again, that could be replaced with a 1664 * efficient population count instruction or similar. 1665 */ 1666 #ifdef CONFIG_DCACHE_WORD_ACCESS 1667 1668 #include <asm/word-at-a-time.h> 1669 1670 #ifdef CONFIG_64BIT 1671 1672 static inline unsigned int fold_hash(unsigned long hash) 1673 { 1674 return hash_64(hash, 32); 1675 } 1676 1677 #else /* 32-bit case */ 1678 1679 #define fold_hash(x) (x) 1680 1681 #endif 1682 1683 unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1684 { 1685 unsigned long a, mask; 1686 unsigned long hash = 0; 1687 1688 for (;;) { 1689 a = load_unaligned_zeropad(name); 1690 if (len < sizeof(unsigned long)) 1691 break; 1692 hash += a; 1693 hash *= 9; 1694 name += sizeof(unsigned long); 1695 len -= sizeof(unsigned long); 1696 if (!len) 1697 goto done; 1698 } 1699 mask = bytemask_from_count(len); 1700 hash += mask & a; 1701 done: 1702 return fold_hash(hash); 1703 } 1704 EXPORT_SYMBOL(full_name_hash); 1705 1706 /* 1707 * Calculate the length and hash of the path component, and 1708 * return the "hash_len" as the result. 1709 */ 1710 static inline u64 hash_name(const char *name) 1711 { 1712 unsigned long a, b, adata, bdata, mask, hash, len; 1713 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1714 1715 hash = a = 0; 1716 len = -sizeof(unsigned long); 1717 do { 1718 hash = (hash + a) * 9; 1719 len += sizeof(unsigned long); 1720 a = load_unaligned_zeropad(name+len); 1721 b = a ^ REPEAT_BYTE('/'); 1722 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1723 1724 adata = prep_zero_mask(a, adata, &constants); 1725 bdata = prep_zero_mask(b, bdata, &constants); 1726 1727 mask = create_zero_mask(adata | bdata); 1728 1729 hash += a & zero_bytemask(mask); 1730 len += find_zero(mask); 1731 return hashlen_create(fold_hash(hash), len); 1732 } 1733 1734 #else 1735 1736 unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1737 { 1738 unsigned long hash = init_name_hash(); 1739 while (len--) 1740 hash = partial_name_hash(*name++, hash); 1741 return end_name_hash(hash); 1742 } 1743 EXPORT_SYMBOL(full_name_hash); 1744 1745 /* 1746 * We know there's a real path component here of at least 1747 * one character. 1748 */ 1749 static inline u64 hash_name(const char *name) 1750 { 1751 unsigned long hash = init_name_hash(); 1752 unsigned long len = 0, c; 1753 1754 c = (unsigned char)*name; 1755 do { 1756 len++; 1757 hash = partial_name_hash(c, hash); 1758 c = (unsigned char)name[len]; 1759 } while (c && c != '/'); 1760 return hashlen_create(end_name_hash(hash), len); 1761 } 1762 1763 #endif 1764 1765 /* 1766 * Name resolution. 1767 * This is the basic name resolution function, turning a pathname into 1768 * the final dentry. We expect 'base' to be positive and a directory. 1769 * 1770 * Returns 0 and nd will have valid dentry and mnt on success. 1771 * Returns error and drops reference to input namei data on failure. 1772 */ 1773 static int link_path_walk(const char *name, struct nameidata *nd) 1774 { 1775 struct path next; 1776 int err; 1777 1778 while (*name=='/') 1779 name++; 1780 if (!*name) 1781 return 0; 1782 1783 /* At this point we know we have a real path component. */ 1784 for(;;) { 1785 u64 hash_len; 1786 int type; 1787 1788 err = may_lookup(nd); 1789 if (err) 1790 break; 1791 1792 hash_len = hash_name(name); 1793 1794 type = LAST_NORM; 1795 if (name[0] == '.') switch (hashlen_len(hash_len)) { 1796 case 2: 1797 if (name[1] == '.') { 1798 type = LAST_DOTDOT; 1799 nd->flags |= LOOKUP_JUMPED; 1800 } 1801 break; 1802 case 1: 1803 type = LAST_DOT; 1804 } 1805 if (likely(type == LAST_NORM)) { 1806 struct dentry *parent = nd->path.dentry; 1807 nd->flags &= ~LOOKUP_JUMPED; 1808 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1809 struct qstr this = { { .hash_len = hash_len }, .name = name }; 1810 err = parent->d_op->d_hash(parent, &this); 1811 if (err < 0) 1812 break; 1813 hash_len = this.hash_len; 1814 name = this.name; 1815 } 1816 } 1817 1818 nd->last.hash_len = hash_len; 1819 nd->last.name = name; 1820 nd->last_type = type; 1821 1822 name += hashlen_len(hash_len); 1823 if (!*name) 1824 return 0; 1825 /* 1826 * If it wasn't NUL, we know it was '/'. Skip that 1827 * slash, and continue until no more slashes. 1828 */ 1829 do { 1830 name++; 1831 } while (unlikely(*name == '/')); 1832 if (!*name) 1833 return 0; 1834 1835 err = walk_component(nd, &next, LOOKUP_FOLLOW); 1836 if (err < 0) 1837 return err; 1838 1839 if (err) { 1840 err = nested_symlink(&next, nd); 1841 if (err) 1842 return err; 1843 } 1844 if (!d_can_lookup(nd->path.dentry)) { 1845 err = -ENOTDIR; 1846 break; 1847 } 1848 } 1849 terminate_walk(nd); 1850 return err; 1851 } 1852 1853 static int path_init(int dfd, const struct filename *name, unsigned int flags, 1854 struct nameidata *nd) 1855 { 1856 int retval = 0; 1857 const char *s = name->name; 1858 1859 nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1860 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 1861 nd->depth = 0; 1862 nd->base = NULL; 1863 if (flags & LOOKUP_ROOT) { 1864 struct dentry *root = nd->root.dentry; 1865 struct inode *inode = root->d_inode; 1866 if (*s) { 1867 if (!d_can_lookup(root)) 1868 return -ENOTDIR; 1869 retval = inode_permission(inode, MAY_EXEC); 1870 if (retval) 1871 return retval; 1872 } 1873 nd->path = nd->root; 1874 nd->inode = inode; 1875 if (flags & LOOKUP_RCU) { 1876 rcu_read_lock(); 1877 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1878 nd->m_seq = read_seqbegin(&mount_lock); 1879 } else { 1880 path_get(&nd->path); 1881 } 1882 goto done; 1883 } 1884 1885 nd->root.mnt = NULL; 1886 1887 nd->m_seq = read_seqbegin(&mount_lock); 1888 if (*s == '/') { 1889 if (flags & LOOKUP_RCU) { 1890 rcu_read_lock(); 1891 nd->seq = set_root_rcu(nd); 1892 } else { 1893 set_root(nd); 1894 path_get(&nd->root); 1895 } 1896 nd->path = nd->root; 1897 } else if (dfd == AT_FDCWD) { 1898 if (flags & LOOKUP_RCU) { 1899 struct fs_struct *fs = current->fs; 1900 unsigned seq; 1901 1902 rcu_read_lock(); 1903 1904 do { 1905 seq = read_seqcount_begin(&fs->seq); 1906 nd->path = fs->pwd; 1907 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1908 } while (read_seqcount_retry(&fs->seq, seq)); 1909 } else { 1910 get_fs_pwd(current->fs, &nd->path); 1911 } 1912 } else { 1913 /* Caller must check execute permissions on the starting path component */ 1914 struct fd f = fdget_raw(dfd); 1915 struct dentry *dentry; 1916 1917 if (!f.file) 1918 return -EBADF; 1919 1920 dentry = f.file->f_path.dentry; 1921 1922 if (*s) { 1923 if (!d_can_lookup(dentry)) { 1924 fdput(f); 1925 return -ENOTDIR; 1926 } 1927 } 1928 1929 nd->path = f.file->f_path; 1930 if (flags & LOOKUP_RCU) { 1931 if (f.flags & FDPUT_FPUT) 1932 nd->base = f.file; 1933 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1934 rcu_read_lock(); 1935 } else { 1936 path_get(&nd->path); 1937 fdput(f); 1938 } 1939 } 1940 1941 nd->inode = nd->path.dentry->d_inode; 1942 if (!(flags & LOOKUP_RCU)) 1943 goto done; 1944 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 1945 goto done; 1946 if (!(nd->flags & LOOKUP_ROOT)) 1947 nd->root.mnt = NULL; 1948 rcu_read_unlock(); 1949 return -ECHILD; 1950 done: 1951 current->total_link_count = 0; 1952 return link_path_walk(s, nd); 1953 } 1954 1955 static void path_cleanup(struct nameidata *nd) 1956 { 1957 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 1958 path_put(&nd->root); 1959 nd->root.mnt = NULL; 1960 } 1961 if (unlikely(nd->base)) 1962 fput(nd->base); 1963 } 1964 1965 static inline int lookup_last(struct nameidata *nd, struct path *path) 1966 { 1967 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1968 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1969 1970 nd->flags &= ~LOOKUP_PARENT; 1971 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1972 } 1973 1974 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1975 static int path_lookupat(int dfd, const struct filename *name, 1976 unsigned int flags, struct nameidata *nd) 1977 { 1978 struct path path; 1979 int err; 1980 1981 /* 1982 * Path walking is largely split up into 2 different synchronisation 1983 * schemes, rcu-walk and ref-walk (explained in 1984 * Documentation/filesystems/path-lookup.txt). These share much of the 1985 * path walk code, but some things particularly setup, cleanup, and 1986 * following mounts are sufficiently divergent that functions are 1987 * duplicated. Typically there is a function foo(), and its RCU 1988 * analogue, foo_rcu(). 1989 * 1990 * -ECHILD is the error number of choice (just to avoid clashes) that 1991 * is returned if some aspect of an rcu-walk fails. Such an error must 1992 * be handled by restarting a traditional ref-walk (which will always 1993 * be able to complete). 1994 */ 1995 err = path_init(dfd, name, flags, nd); 1996 if (!err && !(flags & LOOKUP_PARENT)) { 1997 err = lookup_last(nd, &path); 1998 while (err > 0) { 1999 void *cookie; 2000 struct path link = path; 2001 err = may_follow_link(&link, nd); 2002 if (unlikely(err)) 2003 break; 2004 nd->flags |= LOOKUP_PARENT; 2005 err = follow_link(&link, nd, &cookie); 2006 if (err) 2007 break; 2008 err = lookup_last(nd, &path); 2009 put_link(nd, &link, cookie); 2010 } 2011 } 2012 2013 if (!err) 2014 err = complete_walk(nd); 2015 2016 if (!err && nd->flags & LOOKUP_DIRECTORY) { 2017 if (!d_can_lookup(nd->path.dentry)) { 2018 path_put(&nd->path); 2019 err = -ENOTDIR; 2020 } 2021 } 2022 2023 path_cleanup(nd); 2024 return err; 2025 } 2026 2027 static int filename_lookup(int dfd, struct filename *name, 2028 unsigned int flags, struct nameidata *nd) 2029 { 2030 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 2031 if (unlikely(retval == -ECHILD)) 2032 retval = path_lookupat(dfd, name, flags, nd); 2033 if (unlikely(retval == -ESTALE)) 2034 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 2035 2036 if (likely(!retval)) 2037 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2038 return retval; 2039 } 2040 2041 /* does lookup, returns the object with parent locked */ 2042 struct dentry *kern_path_locked(const char *name, struct path *path) 2043 { 2044 struct filename *filename = getname_kernel(name); 2045 struct nameidata nd; 2046 struct dentry *d; 2047 int err; 2048 2049 if (IS_ERR(filename)) 2050 return ERR_CAST(filename); 2051 2052 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd); 2053 if (err) { 2054 d = ERR_PTR(err); 2055 goto out; 2056 } 2057 if (nd.last_type != LAST_NORM) { 2058 path_put(&nd.path); 2059 d = ERR_PTR(-EINVAL); 2060 goto out; 2061 } 2062 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2063 d = __lookup_hash(&nd.last, nd.path.dentry, 0); 2064 if (IS_ERR(d)) { 2065 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 2066 path_put(&nd.path); 2067 goto out; 2068 } 2069 *path = nd.path; 2070 out: 2071 putname(filename); 2072 return d; 2073 } 2074 2075 int kern_path(const char *name, unsigned int flags, struct path *path) 2076 { 2077 struct nameidata nd; 2078 struct filename *filename = getname_kernel(name); 2079 int res = PTR_ERR(filename); 2080 2081 if (!IS_ERR(filename)) { 2082 res = filename_lookup(AT_FDCWD, filename, flags, &nd); 2083 putname(filename); 2084 if (!res) 2085 *path = nd.path; 2086 } 2087 return res; 2088 } 2089 EXPORT_SYMBOL(kern_path); 2090 2091 /** 2092 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 2093 * @dentry: pointer to dentry of the base directory 2094 * @mnt: pointer to vfs mount of the base directory 2095 * @name: pointer to file name 2096 * @flags: lookup flags 2097 * @path: pointer to struct path to fill 2098 */ 2099 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 2100 const char *name, unsigned int flags, 2101 struct path *path) 2102 { 2103 struct filename *filename = getname_kernel(name); 2104 int err = PTR_ERR(filename); 2105 2106 BUG_ON(flags & LOOKUP_PARENT); 2107 2108 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */ 2109 if (!IS_ERR(filename)) { 2110 struct nameidata nd; 2111 nd.root.dentry = dentry; 2112 nd.root.mnt = mnt; 2113 err = filename_lookup(AT_FDCWD, filename, 2114 flags | LOOKUP_ROOT, &nd); 2115 if (!err) 2116 *path = nd.path; 2117 putname(filename); 2118 } 2119 return err; 2120 } 2121 EXPORT_SYMBOL(vfs_path_lookup); 2122 2123 /* 2124 * Restricted form of lookup. Doesn't follow links, single-component only, 2125 * needs parent already locked. Doesn't follow mounts. 2126 * SMP-safe. 2127 */ 2128 static struct dentry *lookup_hash(struct nameidata *nd) 2129 { 2130 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 2131 } 2132 2133 /** 2134 * lookup_one_len - filesystem helper to lookup single pathname component 2135 * @name: pathname component to lookup 2136 * @base: base directory to lookup from 2137 * @len: maximum length @len should be interpreted to 2138 * 2139 * Note that this routine is purely a helper for filesystem usage and should 2140 * not be called by generic code. 2141 */ 2142 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2143 { 2144 struct qstr this; 2145 unsigned int c; 2146 int err; 2147 2148 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 2149 2150 this.name = name; 2151 this.len = len; 2152 this.hash = full_name_hash(name, len); 2153 if (!len) 2154 return ERR_PTR(-EACCES); 2155 2156 if (unlikely(name[0] == '.')) { 2157 if (len < 2 || (len == 2 && name[1] == '.')) 2158 return ERR_PTR(-EACCES); 2159 } 2160 2161 while (len--) { 2162 c = *(const unsigned char *)name++; 2163 if (c == '/' || c == '\0') 2164 return ERR_PTR(-EACCES); 2165 } 2166 /* 2167 * See if the low-level filesystem might want 2168 * to use its own hash.. 2169 */ 2170 if (base->d_flags & DCACHE_OP_HASH) { 2171 int err = base->d_op->d_hash(base, &this); 2172 if (err < 0) 2173 return ERR_PTR(err); 2174 } 2175 2176 err = inode_permission(base->d_inode, MAY_EXEC); 2177 if (err) 2178 return ERR_PTR(err); 2179 2180 return __lookup_hash(&this, base, 0); 2181 } 2182 EXPORT_SYMBOL(lookup_one_len); 2183 2184 int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 2185 struct path *path, int *empty) 2186 { 2187 struct nameidata nd; 2188 struct filename *tmp = getname_flags(name, flags, empty); 2189 int err = PTR_ERR(tmp); 2190 if (!IS_ERR(tmp)) { 2191 2192 BUG_ON(flags & LOOKUP_PARENT); 2193 2194 err = filename_lookup(dfd, tmp, flags, &nd); 2195 putname(tmp); 2196 if (!err) 2197 *path = nd.path; 2198 } 2199 return err; 2200 } 2201 2202 int user_path_at(int dfd, const char __user *name, unsigned flags, 2203 struct path *path) 2204 { 2205 return user_path_at_empty(dfd, name, flags, path, NULL); 2206 } 2207 EXPORT_SYMBOL(user_path_at); 2208 2209 /* 2210 * NB: most callers don't do anything directly with the reference to the 2211 * to struct filename, but the nd->last pointer points into the name string 2212 * allocated by getname. So we must hold the reference to it until all 2213 * path-walking is complete. 2214 */ 2215 static struct filename * 2216 user_path_parent(int dfd, const char __user *path, struct nameidata *nd, 2217 unsigned int flags) 2218 { 2219 struct filename *s = getname(path); 2220 int error; 2221 2222 /* only LOOKUP_REVAL is allowed in extra flags */ 2223 flags &= LOOKUP_REVAL; 2224 2225 if (IS_ERR(s)) 2226 return s; 2227 2228 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); 2229 if (error) { 2230 putname(s); 2231 return ERR_PTR(error); 2232 } 2233 2234 return s; 2235 } 2236 2237 /** 2238 * mountpoint_last - look up last component for umount 2239 * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 2240 * @path: pointer to container for result 2241 * 2242 * This is a special lookup_last function just for umount. In this case, we 2243 * need to resolve the path without doing any revalidation. 2244 * 2245 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 2246 * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 2247 * in almost all cases, this lookup will be served out of the dcache. The only 2248 * cases where it won't are if nd->last refers to a symlink or the path is 2249 * bogus and it doesn't exist. 2250 * 2251 * Returns: 2252 * -error: if there was an error during lookup. This includes -ENOENT if the 2253 * lookup found a negative dentry. The nd->path reference will also be 2254 * put in this case. 2255 * 2256 * 0: if we successfully resolved nd->path and found it to not to be a 2257 * symlink that needs to be followed. "path" will also be populated. 2258 * The nd->path reference will also be put. 2259 * 2260 * 1: if we successfully resolved nd->last and found it to be a symlink 2261 * that needs to be followed. "path" will be populated with the path 2262 * to the link, and nd->path will *not* be put. 2263 */ 2264 static int 2265 mountpoint_last(struct nameidata *nd, struct path *path) 2266 { 2267 int error = 0; 2268 struct dentry *dentry; 2269 struct dentry *dir = nd->path.dentry; 2270 2271 /* If we're in rcuwalk, drop out of it to handle last component */ 2272 if (nd->flags & LOOKUP_RCU) { 2273 if (unlazy_walk(nd, NULL)) { 2274 error = -ECHILD; 2275 goto out; 2276 } 2277 } 2278 2279 nd->flags &= ~LOOKUP_PARENT; 2280 2281 if (unlikely(nd->last_type != LAST_NORM)) { 2282 error = handle_dots(nd, nd->last_type); 2283 if (error) 2284 goto out; 2285 dentry = dget(nd->path.dentry); 2286 goto done; 2287 } 2288 2289 mutex_lock(&dir->d_inode->i_mutex); 2290 dentry = d_lookup(dir, &nd->last); 2291 if (!dentry) { 2292 /* 2293 * No cached dentry. Mounted dentries are pinned in the cache, 2294 * so that means that this dentry is probably a symlink or the 2295 * path doesn't actually point to a mounted dentry. 2296 */ 2297 dentry = d_alloc(dir, &nd->last); 2298 if (!dentry) { 2299 error = -ENOMEM; 2300 mutex_unlock(&dir->d_inode->i_mutex); 2301 goto out; 2302 } 2303 dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2304 error = PTR_ERR(dentry); 2305 if (IS_ERR(dentry)) { 2306 mutex_unlock(&dir->d_inode->i_mutex); 2307 goto out; 2308 } 2309 } 2310 mutex_unlock(&dir->d_inode->i_mutex); 2311 2312 done: 2313 if (d_is_negative(dentry)) { 2314 error = -ENOENT; 2315 dput(dentry); 2316 goto out; 2317 } 2318 path->dentry = dentry; 2319 path->mnt = nd->path.mnt; 2320 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) 2321 return 1; 2322 mntget(path->mnt); 2323 follow_mount(path); 2324 error = 0; 2325 out: 2326 terminate_walk(nd); 2327 return error; 2328 } 2329 2330 /** 2331 * path_mountpoint - look up a path to be umounted 2332 * @dfd: directory file descriptor to start walk from 2333 * @name: full pathname to walk 2334 * @path: pointer to container for result 2335 * @flags: lookup flags 2336 * 2337 * Look up the given name, but don't attempt to revalidate the last component. 2338 * Returns 0 and "path" will be valid on success; Returns error otherwise. 2339 */ 2340 static int 2341 path_mountpoint(int dfd, const struct filename *name, struct path *path, 2342 unsigned int flags) 2343 { 2344 struct nameidata nd; 2345 int err; 2346 2347 err = path_init(dfd, name, flags, &nd); 2348 if (unlikely(err)) 2349 goto out; 2350 2351 err = mountpoint_last(&nd, path); 2352 while (err > 0) { 2353 void *cookie; 2354 struct path link = *path; 2355 err = may_follow_link(&link, &nd); 2356 if (unlikely(err)) 2357 break; 2358 nd.flags |= LOOKUP_PARENT; 2359 err = follow_link(&link, &nd, &cookie); 2360 if (err) 2361 break; 2362 err = mountpoint_last(&nd, path); 2363 put_link(&nd, &link, cookie); 2364 } 2365 out: 2366 path_cleanup(&nd); 2367 return err; 2368 } 2369 2370 static int 2371 filename_mountpoint(int dfd, struct filename *name, struct path *path, 2372 unsigned int flags) 2373 { 2374 int error; 2375 if (IS_ERR(name)) 2376 return PTR_ERR(name); 2377 error = path_mountpoint(dfd, name, path, flags | LOOKUP_RCU); 2378 if (unlikely(error == -ECHILD)) 2379 error = path_mountpoint(dfd, name, path, flags); 2380 if (unlikely(error == -ESTALE)) 2381 error = path_mountpoint(dfd, name, path, flags | LOOKUP_REVAL); 2382 if (likely(!error)) 2383 audit_inode(name, path->dentry, 0); 2384 putname(name); 2385 return error; 2386 } 2387 2388 /** 2389 * user_path_mountpoint_at - lookup a path from userland in order to umount it 2390 * @dfd: directory file descriptor 2391 * @name: pathname from userland 2392 * @flags: lookup flags 2393 * @path: pointer to container to hold result 2394 * 2395 * A umount is a special case for path walking. We're not actually interested 2396 * in the inode in this situation, and ESTALE errors can be a problem. We 2397 * simply want track down the dentry and vfsmount attached at the mountpoint 2398 * and avoid revalidating the last component. 2399 * 2400 * Returns 0 and populates "path" on success. 2401 */ 2402 int 2403 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 2404 struct path *path) 2405 { 2406 return filename_mountpoint(dfd, getname(name), path, flags); 2407 } 2408 2409 int 2410 kern_path_mountpoint(int dfd, const char *name, struct path *path, 2411 unsigned int flags) 2412 { 2413 return filename_mountpoint(dfd, getname_kernel(name), path, flags); 2414 } 2415 EXPORT_SYMBOL(kern_path_mountpoint); 2416 2417 int __check_sticky(struct inode *dir, struct inode *inode) 2418 { 2419 kuid_t fsuid = current_fsuid(); 2420 2421 if (uid_eq(inode->i_uid, fsuid)) 2422 return 0; 2423 if (uid_eq(dir->i_uid, fsuid)) 2424 return 0; 2425 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 2426 } 2427 EXPORT_SYMBOL(__check_sticky); 2428 2429 /* 2430 * Check whether we can remove a link victim from directory dir, check 2431 * whether the type of victim is right. 2432 * 1. We can't do it if dir is read-only (done in permission()) 2433 * 2. We should have write and exec permissions on dir 2434 * 3. We can't remove anything from append-only dir 2435 * 4. We can't do anything with immutable dir (done in permission()) 2436 * 5. If the sticky bit on dir is set we should either 2437 * a. be owner of dir, or 2438 * b. be owner of victim, or 2439 * c. have CAP_FOWNER capability 2440 * 6. If the victim is append-only or immutable we can't do antyhing with 2441 * links pointing to it. 2442 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 2443 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 2444 * 9. We can't remove a root or mountpoint. 2445 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 2446 * nfs_async_unlink(). 2447 */ 2448 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 2449 { 2450 struct inode *inode = victim->d_inode; 2451 int error; 2452 2453 if (d_is_negative(victim)) 2454 return -ENOENT; 2455 BUG_ON(!inode); 2456 2457 BUG_ON(victim->d_parent->d_inode != dir); 2458 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 2459 2460 error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 2461 if (error) 2462 return error; 2463 if (IS_APPEND(dir)) 2464 return -EPERM; 2465 2466 if (check_sticky(dir, inode) || IS_APPEND(inode) || 2467 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 2468 return -EPERM; 2469 if (isdir) { 2470 if (!d_is_dir(victim)) 2471 return -ENOTDIR; 2472 if (IS_ROOT(victim)) 2473 return -EBUSY; 2474 } else if (d_is_dir(victim)) 2475 return -EISDIR; 2476 if (IS_DEADDIR(dir)) 2477 return -ENOENT; 2478 if (victim->d_flags & DCACHE_NFSFS_RENAMED) 2479 return -EBUSY; 2480 return 0; 2481 } 2482 2483 /* Check whether we can create an object with dentry child in directory 2484 * dir. 2485 * 1. We can't do it if child already exists (open has special treatment for 2486 * this case, but since we are inlined it's OK) 2487 * 2. We can't do it if dir is read-only (done in permission()) 2488 * 3. We should have write and exec permissions on dir 2489 * 4. We can't do it if dir is immutable (done in permission()) 2490 */ 2491 static inline int may_create(struct inode *dir, struct dentry *child) 2492 { 2493 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 2494 if (child->d_inode) 2495 return -EEXIST; 2496 if (IS_DEADDIR(dir)) 2497 return -ENOENT; 2498 return inode_permission(dir, MAY_WRITE | MAY_EXEC); 2499 } 2500 2501 /* 2502 * p1 and p2 should be directories on the same fs. 2503 */ 2504 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 2505 { 2506 struct dentry *p; 2507 2508 if (p1 == p2) { 2509 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2510 return NULL; 2511 } 2512 2513 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 2514 2515 p = d_ancestor(p2, p1); 2516 if (p) { 2517 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2518 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 2519 return p; 2520 } 2521 2522 p = d_ancestor(p1, p2); 2523 if (p) { 2524 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2525 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 2526 return p; 2527 } 2528 2529 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2530 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 2531 return NULL; 2532 } 2533 EXPORT_SYMBOL(lock_rename); 2534 2535 void unlock_rename(struct dentry *p1, struct dentry *p2) 2536 { 2537 mutex_unlock(&p1->d_inode->i_mutex); 2538 if (p1 != p2) { 2539 mutex_unlock(&p2->d_inode->i_mutex); 2540 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 2541 } 2542 } 2543 EXPORT_SYMBOL(unlock_rename); 2544 2545 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2546 bool want_excl) 2547 { 2548 int error = may_create(dir, dentry); 2549 if (error) 2550 return error; 2551 2552 if (!dir->i_op->create) 2553 return -EACCES; /* shouldn't it be ENOSYS? */ 2554 mode &= S_IALLUGO; 2555 mode |= S_IFREG; 2556 error = security_inode_create(dir, dentry, mode); 2557 if (error) 2558 return error; 2559 error = dir->i_op->create(dir, dentry, mode, want_excl); 2560 if (!error) 2561 fsnotify_create(dir, dentry); 2562 return error; 2563 } 2564 EXPORT_SYMBOL(vfs_create); 2565 2566 static int may_open(struct path *path, int acc_mode, int flag) 2567 { 2568 struct dentry *dentry = path->dentry; 2569 struct inode *inode = dentry->d_inode; 2570 int error; 2571 2572 /* O_PATH? */ 2573 if (!acc_mode) 2574 return 0; 2575 2576 if (!inode) 2577 return -ENOENT; 2578 2579 switch (inode->i_mode & S_IFMT) { 2580 case S_IFLNK: 2581 return -ELOOP; 2582 case S_IFDIR: 2583 if (acc_mode & MAY_WRITE) 2584 return -EISDIR; 2585 break; 2586 case S_IFBLK: 2587 case S_IFCHR: 2588 if (path->mnt->mnt_flags & MNT_NODEV) 2589 return -EACCES; 2590 /*FALLTHRU*/ 2591 case S_IFIFO: 2592 case S_IFSOCK: 2593 flag &= ~O_TRUNC; 2594 break; 2595 } 2596 2597 error = inode_permission(inode, acc_mode); 2598 if (error) 2599 return error; 2600 2601 /* 2602 * An append-only file must be opened in append mode for writing. 2603 */ 2604 if (IS_APPEND(inode)) { 2605 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 2606 return -EPERM; 2607 if (flag & O_TRUNC) 2608 return -EPERM; 2609 } 2610 2611 /* O_NOATIME can only be set by the owner or superuser */ 2612 if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 2613 return -EPERM; 2614 2615 return 0; 2616 } 2617 2618 static int handle_truncate(struct file *filp) 2619 { 2620 struct path *path = &filp->f_path; 2621 struct inode *inode = path->dentry->d_inode; 2622 int error = get_write_access(inode); 2623 if (error) 2624 return error; 2625 /* 2626 * Refuse to truncate files with mandatory locks held on them. 2627 */ 2628 error = locks_verify_locked(filp); 2629 if (!error) 2630 error = security_path_truncate(path); 2631 if (!error) { 2632 error = do_truncate(path->dentry, 0, 2633 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2634 filp); 2635 } 2636 put_write_access(inode); 2637 return error; 2638 } 2639 2640 static inline int open_to_namei_flags(int flag) 2641 { 2642 if ((flag & O_ACCMODE) == 3) 2643 flag--; 2644 return flag; 2645 } 2646 2647 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2648 { 2649 int error = security_path_mknod(dir, dentry, mode, 0); 2650 if (error) 2651 return error; 2652 2653 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2654 if (error) 2655 return error; 2656 2657 return security_inode_create(dir->dentry->d_inode, dentry, mode); 2658 } 2659 2660 /* 2661 * Attempt to atomically look up, create and open a file from a negative 2662 * dentry. 2663 * 2664 * Returns 0 if successful. The file will have been created and attached to 2665 * @file by the filesystem calling finish_open(). 2666 * 2667 * Returns 1 if the file was looked up only or didn't need creating. The 2668 * caller will need to perform the open themselves. @path will have been 2669 * updated to point to the new dentry. This may be negative. 2670 * 2671 * Returns an error code otherwise. 2672 */ 2673 static int atomic_open(struct nameidata *nd, struct dentry *dentry, 2674 struct path *path, struct file *file, 2675 const struct open_flags *op, 2676 bool got_write, bool need_lookup, 2677 int *opened) 2678 { 2679 struct inode *dir = nd->path.dentry->d_inode; 2680 unsigned open_flag = open_to_namei_flags(op->open_flag); 2681 umode_t mode; 2682 int error; 2683 int acc_mode; 2684 int create_error = 0; 2685 struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2686 bool excl; 2687 2688 BUG_ON(dentry->d_inode); 2689 2690 /* Don't create child dentry for a dead directory. */ 2691 if (unlikely(IS_DEADDIR(dir))) { 2692 error = -ENOENT; 2693 goto out; 2694 } 2695 2696 mode = op->mode; 2697 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2698 mode &= ~current_umask(); 2699 2700 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2701 if (excl) 2702 open_flag &= ~O_TRUNC; 2703 2704 /* 2705 * Checking write permission is tricky, bacuse we don't know if we are 2706 * going to actually need it: O_CREAT opens should work as long as the 2707 * file exists. But checking existence breaks atomicity. The trick is 2708 * to check access and if not granted clear O_CREAT from the flags. 2709 * 2710 * Another problem is returing the "right" error value (e.g. for an 2711 * O_EXCL open we want to return EEXIST not EROFS). 2712 */ 2713 if (((open_flag & (O_CREAT | O_TRUNC)) || 2714 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 2715 if (!(open_flag & O_CREAT)) { 2716 /* 2717 * No O_CREATE -> atomicity not a requirement -> fall 2718 * back to lookup + open 2719 */ 2720 goto no_open; 2721 } else if (open_flag & (O_EXCL | O_TRUNC)) { 2722 /* Fall back and fail with the right error */ 2723 create_error = -EROFS; 2724 goto no_open; 2725 } else { 2726 /* No side effects, safe to clear O_CREAT */ 2727 create_error = -EROFS; 2728 open_flag &= ~O_CREAT; 2729 } 2730 } 2731 2732 if (open_flag & O_CREAT) { 2733 error = may_o_create(&nd->path, dentry, mode); 2734 if (error) { 2735 create_error = error; 2736 if (open_flag & O_EXCL) 2737 goto no_open; 2738 open_flag &= ~O_CREAT; 2739 } 2740 } 2741 2742 if (nd->flags & LOOKUP_DIRECTORY) 2743 open_flag |= O_DIRECTORY; 2744 2745 file->f_path.dentry = DENTRY_NOT_SET; 2746 file->f_path.mnt = nd->path.mnt; 2747 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 2748 opened); 2749 if (error < 0) { 2750 if (create_error && error == -ENOENT) 2751 error = create_error; 2752 goto out; 2753 } 2754 2755 if (error) { /* returned 1, that is */ 2756 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 2757 error = -EIO; 2758 goto out; 2759 } 2760 if (file->f_path.dentry) { 2761 dput(dentry); 2762 dentry = file->f_path.dentry; 2763 } 2764 if (*opened & FILE_CREATED) 2765 fsnotify_create(dir, dentry); 2766 if (!dentry->d_inode) { 2767 WARN_ON(*opened & FILE_CREATED); 2768 if (create_error) { 2769 error = create_error; 2770 goto out; 2771 } 2772 } else { 2773 if (excl && !(*opened & FILE_CREATED)) { 2774 error = -EEXIST; 2775 goto out; 2776 } 2777 } 2778 goto looked_up; 2779 } 2780 2781 /* 2782 * We didn't have the inode before the open, so check open permission 2783 * here. 2784 */ 2785 acc_mode = op->acc_mode; 2786 if (*opened & FILE_CREATED) { 2787 WARN_ON(!(open_flag & O_CREAT)); 2788 fsnotify_create(dir, dentry); 2789 acc_mode = MAY_OPEN; 2790 } 2791 error = may_open(&file->f_path, acc_mode, open_flag); 2792 if (error) 2793 fput(file); 2794 2795 out: 2796 dput(dentry); 2797 return error; 2798 2799 no_open: 2800 if (need_lookup) { 2801 dentry = lookup_real(dir, dentry, nd->flags); 2802 if (IS_ERR(dentry)) 2803 return PTR_ERR(dentry); 2804 2805 if (create_error) { 2806 int open_flag = op->open_flag; 2807 2808 error = create_error; 2809 if ((open_flag & O_EXCL)) { 2810 if (!dentry->d_inode) 2811 goto out; 2812 } else if (!dentry->d_inode) { 2813 goto out; 2814 } else if ((open_flag & O_TRUNC) && 2815 d_is_reg(dentry)) { 2816 goto out; 2817 } 2818 /* will fail later, go on to get the right error */ 2819 } 2820 } 2821 looked_up: 2822 path->dentry = dentry; 2823 path->mnt = nd->path.mnt; 2824 return 1; 2825 } 2826 2827 /* 2828 * Look up and maybe create and open the last component. 2829 * 2830 * Must be called with i_mutex held on parent. 2831 * 2832 * Returns 0 if the file was successfully atomically created (if necessary) and 2833 * opened. In this case the file will be returned attached to @file. 2834 * 2835 * Returns 1 if the file was not completely opened at this time, though lookups 2836 * and creations will have been performed and the dentry returned in @path will 2837 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 2838 * specified then a negative dentry may be returned. 2839 * 2840 * An error code is returned otherwise. 2841 * 2842 * FILE_CREATE will be set in @*opened if the dentry was created and will be 2843 * cleared otherwise prior to returning. 2844 */ 2845 static int lookup_open(struct nameidata *nd, struct path *path, 2846 struct file *file, 2847 const struct open_flags *op, 2848 bool got_write, int *opened) 2849 { 2850 struct dentry *dir = nd->path.dentry; 2851 struct inode *dir_inode = dir->d_inode; 2852 struct dentry *dentry; 2853 int error; 2854 bool need_lookup; 2855 2856 *opened &= ~FILE_CREATED; 2857 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2858 if (IS_ERR(dentry)) 2859 return PTR_ERR(dentry); 2860 2861 /* Cached positive dentry: will open in f_op->open */ 2862 if (!need_lookup && dentry->d_inode) 2863 goto out_no_open; 2864 2865 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 2866 return atomic_open(nd, dentry, path, file, op, got_write, 2867 need_lookup, opened); 2868 } 2869 2870 if (need_lookup) { 2871 BUG_ON(dentry->d_inode); 2872 2873 dentry = lookup_real(dir_inode, dentry, nd->flags); 2874 if (IS_ERR(dentry)) 2875 return PTR_ERR(dentry); 2876 } 2877 2878 /* Negative dentry, just create the file */ 2879 if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2880 umode_t mode = op->mode; 2881 if (!IS_POSIXACL(dir->d_inode)) 2882 mode &= ~current_umask(); 2883 /* 2884 * This write is needed to ensure that a 2885 * rw->ro transition does not occur between 2886 * the time when the file is created and when 2887 * a permanent write count is taken through 2888 * the 'struct file' in finish_open(). 2889 */ 2890 if (!got_write) { 2891 error = -EROFS; 2892 goto out_dput; 2893 } 2894 *opened |= FILE_CREATED; 2895 error = security_path_mknod(&nd->path, dentry, mode, 0); 2896 if (error) 2897 goto out_dput; 2898 error = vfs_create(dir->d_inode, dentry, mode, 2899 nd->flags & LOOKUP_EXCL); 2900 if (error) 2901 goto out_dput; 2902 } 2903 out_no_open: 2904 path->dentry = dentry; 2905 path->mnt = nd->path.mnt; 2906 return 1; 2907 2908 out_dput: 2909 dput(dentry); 2910 return error; 2911 } 2912 2913 /* 2914 * Handle the last step of open() 2915 */ 2916 static int do_last(struct nameidata *nd, struct path *path, 2917 struct file *file, const struct open_flags *op, 2918 int *opened, struct filename *name) 2919 { 2920 struct dentry *dir = nd->path.dentry; 2921 int open_flag = op->open_flag; 2922 bool will_truncate = (open_flag & O_TRUNC) != 0; 2923 bool got_write = false; 2924 int acc_mode = op->acc_mode; 2925 struct inode *inode; 2926 bool symlink_ok = false; 2927 struct path save_parent = { .dentry = NULL, .mnt = NULL }; 2928 bool retried = false; 2929 int error; 2930 2931 nd->flags &= ~LOOKUP_PARENT; 2932 nd->flags |= op->intent; 2933 2934 if (nd->last_type != LAST_NORM) { 2935 error = handle_dots(nd, nd->last_type); 2936 if (error) 2937 return error; 2938 goto finish_open; 2939 } 2940 2941 if (!(open_flag & O_CREAT)) { 2942 if (nd->last.name[nd->last.len]) 2943 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2944 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 2945 symlink_ok = true; 2946 /* we _can_ be in RCU mode here */ 2947 error = lookup_fast(nd, path, &inode); 2948 if (likely(!error)) 2949 goto finish_lookup; 2950 2951 if (error < 0) 2952 goto out; 2953 2954 BUG_ON(nd->inode != dir->d_inode); 2955 } else { 2956 /* create side of things */ 2957 /* 2958 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2959 * has been cleared when we got to the last component we are 2960 * about to look up 2961 */ 2962 error = complete_walk(nd); 2963 if (error) 2964 return error; 2965 2966 audit_inode(name, dir, LOOKUP_PARENT); 2967 error = -EISDIR; 2968 /* trailing slashes? */ 2969 if (nd->last.name[nd->last.len]) 2970 goto out; 2971 } 2972 2973 retry_lookup: 2974 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 2975 error = mnt_want_write(nd->path.mnt); 2976 if (!error) 2977 got_write = true; 2978 /* 2979 * do _not_ fail yet - we might not need that or fail with 2980 * a different error; let lookup_open() decide; we'll be 2981 * dropping this one anyway. 2982 */ 2983 } 2984 mutex_lock(&dir->d_inode->i_mutex); 2985 error = lookup_open(nd, path, file, op, got_write, opened); 2986 mutex_unlock(&dir->d_inode->i_mutex); 2987 2988 if (error <= 0) { 2989 if (error) 2990 goto out; 2991 2992 if ((*opened & FILE_CREATED) || 2993 !S_ISREG(file_inode(file)->i_mode)) 2994 will_truncate = false; 2995 2996 audit_inode(name, file->f_path.dentry, 0); 2997 goto opened; 2998 } 2999 3000 if (*opened & FILE_CREATED) { 3001 /* Don't check for write permission, don't truncate */ 3002 open_flag &= ~O_TRUNC; 3003 will_truncate = false; 3004 acc_mode = MAY_OPEN; 3005 path_to_nameidata(path, nd); 3006 goto finish_open_created; 3007 } 3008 3009 /* 3010 * create/update audit record if it already exists. 3011 */ 3012 if (d_is_positive(path->dentry)) 3013 audit_inode(name, path->dentry, 0); 3014 3015 /* 3016 * If atomic_open() acquired write access it is dropped now due to 3017 * possible mount and symlink following (this might be optimized away if 3018 * necessary...) 3019 */ 3020 if (got_write) { 3021 mnt_drop_write(nd->path.mnt); 3022 got_write = false; 3023 } 3024 3025 error = -EEXIST; 3026 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 3027 goto exit_dput; 3028 3029 error = follow_managed(path, nd->flags); 3030 if (error < 0) 3031 goto exit_dput; 3032 3033 if (error) 3034 nd->flags |= LOOKUP_JUMPED; 3035 3036 BUG_ON(nd->flags & LOOKUP_RCU); 3037 inode = path->dentry->d_inode; 3038 finish_lookup: 3039 /* we _can_ be in RCU mode here */ 3040 error = -ENOENT; 3041 if (d_is_negative(path->dentry)) { 3042 path_to_nameidata(path, nd); 3043 goto out; 3044 } 3045 3046 if (should_follow_link(path->dentry, !symlink_ok)) { 3047 if (nd->flags & LOOKUP_RCU) { 3048 if (unlikely(unlazy_walk(nd, path->dentry))) { 3049 error = -ECHILD; 3050 goto out; 3051 } 3052 } 3053 BUG_ON(inode != path->dentry->d_inode); 3054 return 1; 3055 } 3056 3057 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 3058 path_to_nameidata(path, nd); 3059 } else { 3060 save_parent.dentry = nd->path.dentry; 3061 save_parent.mnt = mntget(path->mnt); 3062 nd->path.dentry = path->dentry; 3063 3064 } 3065 nd->inode = inode; 3066 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3067 finish_open: 3068 error = complete_walk(nd); 3069 if (error) { 3070 path_put(&save_parent); 3071 return error; 3072 } 3073 audit_inode(name, nd->path.dentry, 0); 3074 error = -EISDIR; 3075 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 3076 goto out; 3077 error = -ENOTDIR; 3078 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3079 goto out; 3080 if (!d_is_reg(nd->path.dentry)) 3081 will_truncate = false; 3082 3083 if (will_truncate) { 3084 error = mnt_want_write(nd->path.mnt); 3085 if (error) 3086 goto out; 3087 got_write = true; 3088 } 3089 finish_open_created: 3090 error = may_open(&nd->path, acc_mode, open_flag); 3091 if (error) 3092 goto out; 3093 3094 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 3095 error = vfs_open(&nd->path, file, current_cred()); 3096 if (!error) { 3097 *opened |= FILE_OPENED; 3098 } else { 3099 if (error == -EOPENSTALE) 3100 goto stale_open; 3101 goto out; 3102 } 3103 opened: 3104 error = open_check_o_direct(file); 3105 if (error) 3106 goto exit_fput; 3107 error = ima_file_check(file, op->acc_mode, *opened); 3108 if (error) 3109 goto exit_fput; 3110 3111 if (will_truncate) { 3112 error = handle_truncate(file); 3113 if (error) 3114 goto exit_fput; 3115 } 3116 out: 3117 if (got_write) 3118 mnt_drop_write(nd->path.mnt); 3119 path_put(&save_parent); 3120 terminate_walk(nd); 3121 return error; 3122 3123 exit_dput: 3124 path_put_conditional(path, nd); 3125 goto out; 3126 exit_fput: 3127 fput(file); 3128 goto out; 3129 3130 stale_open: 3131 /* If no saved parent or already retried then can't retry */ 3132 if (!save_parent.dentry || retried) 3133 goto out; 3134 3135 BUG_ON(save_parent.dentry != dir); 3136 path_put(&nd->path); 3137 nd->path = save_parent; 3138 nd->inode = dir->d_inode; 3139 save_parent.mnt = NULL; 3140 save_parent.dentry = NULL; 3141 if (got_write) { 3142 mnt_drop_write(nd->path.mnt); 3143 got_write = false; 3144 } 3145 retried = true; 3146 goto retry_lookup; 3147 } 3148 3149 static int do_tmpfile(int dfd, struct filename *pathname, 3150 struct nameidata *nd, int flags, 3151 const struct open_flags *op, 3152 struct file *file, int *opened) 3153 { 3154 static const struct qstr name = QSTR_INIT("/", 1); 3155 struct dentry *dentry, *child; 3156 struct inode *dir; 3157 int error = path_lookupat(dfd, pathname, 3158 flags | LOOKUP_DIRECTORY, nd); 3159 if (unlikely(error)) 3160 return error; 3161 error = mnt_want_write(nd->path.mnt); 3162 if (unlikely(error)) 3163 goto out; 3164 /* we want directory to be writable */ 3165 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 3166 if (error) 3167 goto out2; 3168 dentry = nd->path.dentry; 3169 dir = dentry->d_inode; 3170 if (!dir->i_op->tmpfile) { 3171 error = -EOPNOTSUPP; 3172 goto out2; 3173 } 3174 child = d_alloc(dentry, &name); 3175 if (unlikely(!child)) { 3176 error = -ENOMEM; 3177 goto out2; 3178 } 3179 nd->flags &= ~LOOKUP_DIRECTORY; 3180 nd->flags |= op->intent; 3181 dput(nd->path.dentry); 3182 nd->path.dentry = child; 3183 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 3184 if (error) 3185 goto out2; 3186 audit_inode(pathname, nd->path.dentry, 0); 3187 /* Don't check for other permissions, the inode was just created */ 3188 error = may_open(&nd->path, MAY_OPEN, op->open_flag); 3189 if (error) 3190 goto out2; 3191 file->f_path.mnt = nd->path.mnt; 3192 error = finish_open(file, nd->path.dentry, NULL, opened); 3193 if (error) 3194 goto out2; 3195 error = open_check_o_direct(file); 3196 if (error) { 3197 fput(file); 3198 } else if (!(op->open_flag & O_EXCL)) { 3199 struct inode *inode = file_inode(file); 3200 spin_lock(&inode->i_lock); 3201 inode->i_state |= I_LINKABLE; 3202 spin_unlock(&inode->i_lock); 3203 } 3204 out2: 3205 mnt_drop_write(nd->path.mnt); 3206 out: 3207 path_put(&nd->path); 3208 return error; 3209 } 3210 3211 static struct file *path_openat(int dfd, struct filename *pathname, 3212 struct nameidata *nd, const struct open_flags *op, int flags) 3213 { 3214 struct file *file; 3215 struct path path; 3216 int opened = 0; 3217 int error; 3218 3219 file = get_empty_filp(); 3220 if (IS_ERR(file)) 3221 return file; 3222 3223 file->f_flags = op->open_flag; 3224 3225 if (unlikely(file->f_flags & __O_TMPFILE)) { 3226 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 3227 goto out; 3228 } 3229 3230 error = path_init(dfd, pathname, flags, nd); 3231 if (unlikely(error)) 3232 goto out; 3233 3234 error = do_last(nd, &path, file, op, &opened, pathname); 3235 while (unlikely(error > 0)) { /* trailing symlink */ 3236 struct path link = path; 3237 void *cookie; 3238 if (!(nd->flags & LOOKUP_FOLLOW)) { 3239 path_put_conditional(&path, nd); 3240 path_put(&nd->path); 3241 error = -ELOOP; 3242 break; 3243 } 3244 error = may_follow_link(&link, nd); 3245 if (unlikely(error)) 3246 break; 3247 nd->flags |= LOOKUP_PARENT; 3248 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3249 error = follow_link(&link, nd, &cookie); 3250 if (unlikely(error)) 3251 break; 3252 error = do_last(nd, &path, file, op, &opened, pathname); 3253 put_link(nd, &link, cookie); 3254 } 3255 out: 3256 path_cleanup(nd); 3257 if (!(opened & FILE_OPENED)) { 3258 BUG_ON(!error); 3259 put_filp(file); 3260 } 3261 if (unlikely(error)) { 3262 if (error == -EOPENSTALE) { 3263 if (flags & LOOKUP_RCU) 3264 error = -ECHILD; 3265 else 3266 error = -ESTALE; 3267 } 3268 file = ERR_PTR(error); 3269 } 3270 return file; 3271 } 3272 3273 struct file *do_filp_open(int dfd, struct filename *pathname, 3274 const struct open_flags *op) 3275 { 3276 struct nameidata nd; 3277 int flags = op->lookup_flags; 3278 struct file *filp; 3279 3280 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 3281 if (unlikely(filp == ERR_PTR(-ECHILD))) 3282 filp = path_openat(dfd, pathname, &nd, op, flags); 3283 if (unlikely(filp == ERR_PTR(-ESTALE))) 3284 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 3285 return filp; 3286 } 3287 3288 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3289 const char *name, const struct open_flags *op) 3290 { 3291 struct nameidata nd; 3292 struct file *file; 3293 struct filename *filename; 3294 int flags = op->lookup_flags | LOOKUP_ROOT; 3295 3296 nd.root.mnt = mnt; 3297 nd.root.dentry = dentry; 3298 3299 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 3300 return ERR_PTR(-ELOOP); 3301 3302 filename = getname_kernel(name); 3303 if (unlikely(IS_ERR(filename))) 3304 return ERR_CAST(filename); 3305 3306 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU); 3307 if (unlikely(file == ERR_PTR(-ECHILD))) 3308 file = path_openat(-1, filename, &nd, op, flags); 3309 if (unlikely(file == ERR_PTR(-ESTALE))) 3310 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL); 3311 putname(filename); 3312 return file; 3313 } 3314 3315 static struct dentry *filename_create(int dfd, struct filename *name, 3316 struct path *path, unsigned int lookup_flags) 3317 { 3318 struct dentry *dentry = ERR_PTR(-EEXIST); 3319 struct nameidata nd; 3320 int err2; 3321 int error; 3322 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 3323 3324 /* 3325 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 3326 * other flags passed in are ignored! 3327 */ 3328 lookup_flags &= LOOKUP_REVAL; 3329 3330 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd); 3331 if (error) 3332 return ERR_PTR(error); 3333 3334 /* 3335 * Yucky last component or no last component at all? 3336 * (foo/., foo/.., /////) 3337 */ 3338 if (nd.last_type != LAST_NORM) 3339 goto out; 3340 nd.flags &= ~LOOKUP_PARENT; 3341 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3342 3343 /* don't fail immediately if it's r/o, at least try to report other errors */ 3344 err2 = mnt_want_write(nd.path.mnt); 3345 /* 3346 * Do the final lookup. 3347 */ 3348 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3349 dentry = lookup_hash(&nd); 3350 if (IS_ERR(dentry)) 3351 goto unlock; 3352 3353 error = -EEXIST; 3354 if (d_is_positive(dentry)) 3355 goto fail; 3356 3357 /* 3358 * Special case - lookup gave negative, but... we had foo/bar/ 3359 * From the vfs_mknod() POV we just have a negative dentry - 3360 * all is fine. Let's be bastards - you had / on the end, you've 3361 * been asking for (non-existent) directory. -ENOENT for you. 3362 */ 3363 if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3364 error = -ENOENT; 3365 goto fail; 3366 } 3367 if (unlikely(err2)) { 3368 error = err2; 3369 goto fail; 3370 } 3371 *path = nd.path; 3372 return dentry; 3373 fail: 3374 dput(dentry); 3375 dentry = ERR_PTR(error); 3376 unlock: 3377 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3378 if (!err2) 3379 mnt_drop_write(nd.path.mnt); 3380 out: 3381 path_put(&nd.path); 3382 return dentry; 3383 } 3384 3385 struct dentry *kern_path_create(int dfd, const char *pathname, 3386 struct path *path, unsigned int lookup_flags) 3387 { 3388 struct filename *filename = getname_kernel(pathname); 3389 struct dentry *res; 3390 3391 if (IS_ERR(filename)) 3392 return ERR_CAST(filename); 3393 res = filename_create(dfd, filename, path, lookup_flags); 3394 putname(filename); 3395 return res; 3396 } 3397 EXPORT_SYMBOL(kern_path_create); 3398 3399 void done_path_create(struct path *path, struct dentry *dentry) 3400 { 3401 dput(dentry); 3402 mutex_unlock(&path->dentry->d_inode->i_mutex); 3403 mnt_drop_write(path->mnt); 3404 path_put(path); 3405 } 3406 EXPORT_SYMBOL(done_path_create); 3407 3408 struct dentry *user_path_create(int dfd, const char __user *pathname, 3409 struct path *path, unsigned int lookup_flags) 3410 { 3411 struct filename *tmp = getname(pathname); 3412 struct dentry *res; 3413 if (IS_ERR(tmp)) 3414 return ERR_CAST(tmp); 3415 res = filename_create(dfd, tmp, path, lookup_flags); 3416 putname(tmp); 3417 return res; 3418 } 3419 EXPORT_SYMBOL(user_path_create); 3420 3421 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 3422 { 3423 int error = may_create(dir, dentry); 3424 3425 if (error) 3426 return error; 3427 3428 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 3429 return -EPERM; 3430 3431 if (!dir->i_op->mknod) 3432 return -EPERM; 3433 3434 error = devcgroup_inode_mknod(mode, dev); 3435 if (error) 3436 return error; 3437 3438 error = security_inode_mknod(dir, dentry, mode, dev); 3439 if (error) 3440 return error; 3441 3442 error = dir->i_op->mknod(dir, dentry, mode, dev); 3443 if (!error) 3444 fsnotify_create(dir, dentry); 3445 return error; 3446 } 3447 EXPORT_SYMBOL(vfs_mknod); 3448 3449 static int may_mknod(umode_t mode) 3450 { 3451 switch (mode & S_IFMT) { 3452 case S_IFREG: 3453 case S_IFCHR: 3454 case S_IFBLK: 3455 case S_IFIFO: 3456 case S_IFSOCK: 3457 case 0: /* zero mode translates to S_IFREG */ 3458 return 0; 3459 case S_IFDIR: 3460 return -EPERM; 3461 default: 3462 return -EINVAL; 3463 } 3464 } 3465 3466 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 3467 unsigned, dev) 3468 { 3469 struct dentry *dentry; 3470 struct path path; 3471 int error; 3472 unsigned int lookup_flags = 0; 3473 3474 error = may_mknod(mode); 3475 if (error) 3476 return error; 3477 retry: 3478 dentry = user_path_create(dfd, filename, &path, lookup_flags); 3479 if (IS_ERR(dentry)) 3480 return PTR_ERR(dentry); 3481 3482 if (!IS_POSIXACL(path.dentry->d_inode)) 3483 mode &= ~current_umask(); 3484 error = security_path_mknod(&path, dentry, mode, dev); 3485 if (error) 3486 goto out; 3487 switch (mode & S_IFMT) { 3488 case 0: case S_IFREG: 3489 error = vfs_create(path.dentry->d_inode,dentry,mode,true); 3490 break; 3491 case S_IFCHR: case S_IFBLK: 3492 error = vfs_mknod(path.dentry->d_inode,dentry,mode, 3493 new_decode_dev(dev)); 3494 break; 3495 case S_IFIFO: case S_IFSOCK: 3496 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 3497 break; 3498 } 3499 out: 3500 done_path_create(&path, dentry); 3501 if (retry_estale(error, lookup_flags)) { 3502 lookup_flags |= LOOKUP_REVAL; 3503 goto retry; 3504 } 3505 return error; 3506 } 3507 3508 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 3509 { 3510 return sys_mknodat(AT_FDCWD, filename, mode, dev); 3511 } 3512 3513 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 3514 { 3515 int error = may_create(dir, dentry); 3516 unsigned max_links = dir->i_sb->s_max_links; 3517 3518 if (error) 3519 return error; 3520 3521 if (!dir->i_op->mkdir) 3522 return -EPERM; 3523 3524 mode &= (S_IRWXUGO|S_ISVTX); 3525 error = security_inode_mkdir(dir, dentry, mode); 3526 if (error) 3527 return error; 3528 3529 if (max_links && dir->i_nlink >= max_links) 3530 return -EMLINK; 3531 3532 error = dir->i_op->mkdir(dir, dentry, mode); 3533 if (!error) 3534 fsnotify_mkdir(dir, dentry); 3535 return error; 3536 } 3537 EXPORT_SYMBOL(vfs_mkdir); 3538 3539 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 3540 { 3541 struct dentry *dentry; 3542 struct path path; 3543 int error; 3544 unsigned int lookup_flags = LOOKUP_DIRECTORY; 3545 3546 retry: 3547 dentry = user_path_create(dfd, pathname, &path, lookup_flags); 3548 if (IS_ERR(dentry)) 3549 return PTR_ERR(dentry); 3550 3551 if (!IS_POSIXACL(path.dentry->d_inode)) 3552 mode &= ~current_umask(); 3553 error = security_path_mkdir(&path, dentry, mode); 3554 if (!error) 3555 error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3556 done_path_create(&path, dentry); 3557 if (retry_estale(error, lookup_flags)) { 3558 lookup_flags |= LOOKUP_REVAL; 3559 goto retry; 3560 } 3561 return error; 3562 } 3563 3564 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 3565 { 3566 return sys_mkdirat(AT_FDCWD, pathname, mode); 3567 } 3568 3569 /* 3570 * The dentry_unhash() helper will try to drop the dentry early: we 3571 * should have a usage count of 1 if we're the only user of this 3572 * dentry, and if that is true (possibly after pruning the dcache), 3573 * then we drop the dentry now. 3574 * 3575 * A low-level filesystem can, if it choses, legally 3576 * do a 3577 * 3578 * if (!d_unhashed(dentry)) 3579 * return -EBUSY; 3580 * 3581 * if it cannot handle the case of removing a directory 3582 * that is still in use by something else.. 3583 */ 3584 void dentry_unhash(struct dentry *dentry) 3585 { 3586 shrink_dcache_parent(dentry); 3587 spin_lock(&dentry->d_lock); 3588 if (dentry->d_lockref.count == 1) 3589 __d_drop(dentry); 3590 spin_unlock(&dentry->d_lock); 3591 } 3592 EXPORT_SYMBOL(dentry_unhash); 3593 3594 int vfs_rmdir(struct inode *dir, struct dentry *dentry) 3595 { 3596 int error = may_delete(dir, dentry, 1); 3597 3598 if (error) 3599 return error; 3600 3601 if (!dir->i_op->rmdir) 3602 return -EPERM; 3603 3604 dget(dentry); 3605 mutex_lock(&dentry->d_inode->i_mutex); 3606 3607 error = -EBUSY; 3608 if (is_local_mountpoint(dentry)) 3609 goto out; 3610 3611 error = security_inode_rmdir(dir, dentry); 3612 if (error) 3613 goto out; 3614 3615 shrink_dcache_parent(dentry); 3616 error = dir->i_op->rmdir(dir, dentry); 3617 if (error) 3618 goto out; 3619 3620 dentry->d_inode->i_flags |= S_DEAD; 3621 dont_mount(dentry); 3622 detach_mounts(dentry); 3623 3624 out: 3625 mutex_unlock(&dentry->d_inode->i_mutex); 3626 dput(dentry); 3627 if (!error) 3628 d_delete(dentry); 3629 return error; 3630 } 3631 EXPORT_SYMBOL(vfs_rmdir); 3632 3633 static long do_rmdir(int dfd, const char __user *pathname) 3634 { 3635 int error = 0; 3636 struct filename *name; 3637 struct dentry *dentry; 3638 struct nameidata nd; 3639 unsigned int lookup_flags = 0; 3640 retry: 3641 name = user_path_parent(dfd, pathname, &nd, lookup_flags); 3642 if (IS_ERR(name)) 3643 return PTR_ERR(name); 3644 3645 switch(nd.last_type) { 3646 case LAST_DOTDOT: 3647 error = -ENOTEMPTY; 3648 goto exit1; 3649 case LAST_DOT: 3650 error = -EINVAL; 3651 goto exit1; 3652 case LAST_ROOT: 3653 error = -EBUSY; 3654 goto exit1; 3655 } 3656 3657 nd.flags &= ~LOOKUP_PARENT; 3658 error = mnt_want_write(nd.path.mnt); 3659 if (error) 3660 goto exit1; 3661 3662 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3663 dentry = lookup_hash(&nd); 3664 error = PTR_ERR(dentry); 3665 if (IS_ERR(dentry)) 3666 goto exit2; 3667 if (!dentry->d_inode) { 3668 error = -ENOENT; 3669 goto exit3; 3670 } 3671 error = security_path_rmdir(&nd.path, dentry); 3672 if (error) 3673 goto exit3; 3674 error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 3675 exit3: 3676 dput(dentry); 3677 exit2: 3678 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3679 mnt_drop_write(nd.path.mnt); 3680 exit1: 3681 path_put(&nd.path); 3682 putname(name); 3683 if (retry_estale(error, lookup_flags)) { 3684 lookup_flags |= LOOKUP_REVAL; 3685 goto retry; 3686 } 3687 return error; 3688 } 3689 3690 SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 3691 { 3692 return do_rmdir(AT_FDCWD, pathname); 3693 } 3694 3695 /** 3696 * vfs_unlink - unlink a filesystem object 3697 * @dir: parent directory 3698 * @dentry: victim 3699 * @delegated_inode: returns victim inode, if the inode is delegated. 3700 * 3701 * The caller must hold dir->i_mutex. 3702 * 3703 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3704 * return a reference to the inode in delegated_inode. The caller 3705 * should then break the delegation on that inode and retry. Because 3706 * breaking a delegation may take a long time, the caller should drop 3707 * dir->i_mutex before doing so. 3708 * 3709 * Alternatively, a caller may pass NULL for delegated_inode. This may 3710 * be appropriate for callers that expect the underlying filesystem not 3711 * to be NFS exported. 3712 */ 3713 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 3714 { 3715 struct inode *target = dentry->d_inode; 3716 int error = may_delete(dir, dentry, 0); 3717 3718 if (error) 3719 return error; 3720 3721 if (!dir->i_op->unlink) 3722 return -EPERM; 3723 3724 mutex_lock(&target->i_mutex); 3725 if (is_local_mountpoint(dentry)) 3726 error = -EBUSY; 3727 else { 3728 error = security_inode_unlink(dir, dentry); 3729 if (!error) { 3730 error = try_break_deleg(target, delegated_inode); 3731 if (error) 3732 goto out; 3733 error = dir->i_op->unlink(dir, dentry); 3734 if (!error) { 3735 dont_mount(dentry); 3736 detach_mounts(dentry); 3737 } 3738 } 3739 } 3740 out: 3741 mutex_unlock(&target->i_mutex); 3742 3743 /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 3744 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 3745 fsnotify_link_count(target); 3746 d_delete(dentry); 3747 } 3748 3749 return error; 3750 } 3751 EXPORT_SYMBOL(vfs_unlink); 3752 3753 /* 3754 * Make sure that the actual truncation of the file will occur outside its 3755 * directory's i_mutex. Truncate can take a long time if there is a lot of 3756 * writeout happening, and we don't want to prevent access to the directory 3757 * while waiting on the I/O. 3758 */ 3759 static long do_unlinkat(int dfd, const char __user *pathname) 3760 { 3761 int error; 3762 struct filename *name; 3763 struct dentry *dentry; 3764 struct nameidata nd; 3765 struct inode *inode = NULL; 3766 struct inode *delegated_inode = NULL; 3767 unsigned int lookup_flags = 0; 3768 retry: 3769 name = user_path_parent(dfd, pathname, &nd, lookup_flags); 3770 if (IS_ERR(name)) 3771 return PTR_ERR(name); 3772 3773 error = -EISDIR; 3774 if (nd.last_type != LAST_NORM) 3775 goto exit1; 3776 3777 nd.flags &= ~LOOKUP_PARENT; 3778 error = mnt_want_write(nd.path.mnt); 3779 if (error) 3780 goto exit1; 3781 retry_deleg: 3782 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3783 dentry = lookup_hash(&nd); 3784 error = PTR_ERR(dentry); 3785 if (!IS_ERR(dentry)) { 3786 /* Why not before? Because we want correct error value */ 3787 if (nd.last.name[nd.last.len]) 3788 goto slashes; 3789 inode = dentry->d_inode; 3790 if (d_is_negative(dentry)) 3791 goto slashes; 3792 ihold(inode); 3793 error = security_path_unlink(&nd.path, dentry); 3794 if (error) 3795 goto exit2; 3796 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); 3797 exit2: 3798 dput(dentry); 3799 } 3800 mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3801 if (inode) 3802 iput(inode); /* truncate the inode here */ 3803 inode = NULL; 3804 if (delegated_inode) { 3805 error = break_deleg_wait(&delegated_inode); 3806 if (!error) 3807 goto retry_deleg; 3808 } 3809 mnt_drop_write(nd.path.mnt); 3810 exit1: 3811 path_put(&nd.path); 3812 putname(name); 3813 if (retry_estale(error, lookup_flags)) { 3814 lookup_flags |= LOOKUP_REVAL; 3815 inode = NULL; 3816 goto retry; 3817 } 3818 return error; 3819 3820 slashes: 3821 if (d_is_negative(dentry)) 3822 error = -ENOENT; 3823 else if (d_is_dir(dentry)) 3824 error = -EISDIR; 3825 else 3826 error = -ENOTDIR; 3827 goto exit2; 3828 } 3829 3830 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 3831 { 3832 if ((flag & ~AT_REMOVEDIR) != 0) 3833 return -EINVAL; 3834 3835 if (flag & AT_REMOVEDIR) 3836 return do_rmdir(dfd, pathname); 3837 3838 return do_unlinkat(dfd, pathname); 3839 } 3840 3841 SYSCALL_DEFINE1(unlink, const char __user *, pathname) 3842 { 3843 return do_unlinkat(AT_FDCWD, pathname); 3844 } 3845 3846 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 3847 { 3848 int error = may_create(dir, dentry); 3849 3850 if (error) 3851 return error; 3852 3853 if (!dir->i_op->symlink) 3854 return -EPERM; 3855 3856 error = security_inode_symlink(dir, dentry, oldname); 3857 if (error) 3858 return error; 3859 3860 error = dir->i_op->symlink(dir, dentry, oldname); 3861 if (!error) 3862 fsnotify_create(dir, dentry); 3863 return error; 3864 } 3865 EXPORT_SYMBOL(vfs_symlink); 3866 3867 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 3868 int, newdfd, const char __user *, newname) 3869 { 3870 int error; 3871 struct filename *from; 3872 struct dentry *dentry; 3873 struct path path; 3874 unsigned int lookup_flags = 0; 3875 3876 from = getname(oldname); 3877 if (IS_ERR(from)) 3878 return PTR_ERR(from); 3879 retry: 3880 dentry = user_path_create(newdfd, newname, &path, lookup_flags); 3881 error = PTR_ERR(dentry); 3882 if (IS_ERR(dentry)) 3883 goto out_putname; 3884 3885 error = security_path_symlink(&path, dentry, from->name); 3886 if (!error) 3887 error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3888 done_path_create(&path, dentry); 3889 if (retry_estale(error, lookup_flags)) { 3890 lookup_flags |= LOOKUP_REVAL; 3891 goto retry; 3892 } 3893 out_putname: 3894 putname(from); 3895 return error; 3896 } 3897 3898 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 3899 { 3900 return sys_symlinkat(oldname, AT_FDCWD, newname); 3901 } 3902 3903 /** 3904 * vfs_link - create a new link 3905 * @old_dentry: object to be linked 3906 * @dir: new parent 3907 * @new_dentry: where to create the new link 3908 * @delegated_inode: returns inode needing a delegation break 3909 * 3910 * The caller must hold dir->i_mutex 3911 * 3912 * If vfs_link discovers a delegation on the to-be-linked file in need 3913 * of breaking, it will return -EWOULDBLOCK and return a reference to the 3914 * inode in delegated_inode. The caller should then break the delegation 3915 * and retry. Because breaking a delegation may take a long time, the 3916 * caller should drop the i_mutex before doing so. 3917 * 3918 * Alternatively, a caller may pass NULL for delegated_inode. This may 3919 * be appropriate for callers that expect the underlying filesystem not 3920 * to be NFS exported. 3921 */ 3922 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 3923 { 3924 struct inode *inode = old_dentry->d_inode; 3925 unsigned max_links = dir->i_sb->s_max_links; 3926 int error; 3927 3928 if (!inode) 3929 return -ENOENT; 3930 3931 error = may_create(dir, new_dentry); 3932 if (error) 3933 return error; 3934 3935 if (dir->i_sb != inode->i_sb) 3936 return -EXDEV; 3937 3938 /* 3939 * A link to an append-only or immutable file cannot be created. 3940 */ 3941 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 3942 return -EPERM; 3943 if (!dir->i_op->link) 3944 return -EPERM; 3945 if (S_ISDIR(inode->i_mode)) 3946 return -EPERM; 3947 3948 error = security_inode_link(old_dentry, dir, new_dentry); 3949 if (error) 3950 return error; 3951 3952 mutex_lock(&inode->i_mutex); 3953 /* Make sure we don't allow creating hardlink to an unlinked file */ 3954 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3955 error = -ENOENT; 3956 else if (max_links && inode->i_nlink >= max_links) 3957 error = -EMLINK; 3958 else { 3959 error = try_break_deleg(inode, delegated_inode); 3960 if (!error) 3961 error = dir->i_op->link(old_dentry, dir, new_dentry); 3962 } 3963 3964 if (!error && (inode->i_state & I_LINKABLE)) { 3965 spin_lock(&inode->i_lock); 3966 inode->i_state &= ~I_LINKABLE; 3967 spin_unlock(&inode->i_lock); 3968 } 3969 mutex_unlock(&inode->i_mutex); 3970 if (!error) 3971 fsnotify_link(dir, inode, new_dentry); 3972 return error; 3973 } 3974 EXPORT_SYMBOL(vfs_link); 3975 3976 /* 3977 * Hardlinks are often used in delicate situations. We avoid 3978 * security-related surprises by not following symlinks on the 3979 * newname. --KAB 3980 * 3981 * We don't follow them on the oldname either to be compatible 3982 * with linux 2.0, and to avoid hard-linking to directories 3983 * and other special files. --ADM 3984 */ 3985 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 3986 int, newdfd, const char __user *, newname, int, flags) 3987 { 3988 struct dentry *new_dentry; 3989 struct path old_path, new_path; 3990 struct inode *delegated_inode = NULL; 3991 int how = 0; 3992 int error; 3993 3994 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3995 return -EINVAL; 3996 /* 3997 * To use null names we require CAP_DAC_READ_SEARCH 3998 * This ensures that not everyone will be able to create 3999 * handlink using the passed filedescriptor. 4000 */ 4001 if (flags & AT_EMPTY_PATH) { 4002 if (!capable(CAP_DAC_READ_SEARCH)) 4003 return -ENOENT; 4004 how = LOOKUP_EMPTY; 4005 } 4006 4007 if (flags & AT_SYMLINK_FOLLOW) 4008 how |= LOOKUP_FOLLOW; 4009 retry: 4010 error = user_path_at(olddfd, oldname, how, &old_path); 4011 if (error) 4012 return error; 4013 4014 new_dentry = user_path_create(newdfd, newname, &new_path, 4015 (how & LOOKUP_REVAL)); 4016 error = PTR_ERR(new_dentry); 4017 if (IS_ERR(new_dentry)) 4018 goto out; 4019 4020 error = -EXDEV; 4021 if (old_path.mnt != new_path.mnt) 4022 goto out_dput; 4023 error = may_linkat(&old_path); 4024 if (unlikely(error)) 4025 goto out_dput; 4026 error = security_path_link(old_path.dentry, &new_path, new_dentry); 4027 if (error) 4028 goto out_dput; 4029 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 4030 out_dput: 4031 done_path_create(&new_path, new_dentry); 4032 if (delegated_inode) { 4033 error = break_deleg_wait(&delegated_inode); 4034 if (!error) { 4035 path_put(&old_path); 4036 goto retry; 4037 } 4038 } 4039 if (retry_estale(error, how)) { 4040 path_put(&old_path); 4041 how |= LOOKUP_REVAL; 4042 goto retry; 4043 } 4044 out: 4045 path_put(&old_path); 4046 4047 return error; 4048 } 4049 4050 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 4051 { 4052 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 4053 } 4054 4055 /** 4056 * vfs_rename - rename a filesystem object 4057 * @old_dir: parent of source 4058 * @old_dentry: source 4059 * @new_dir: parent of destination 4060 * @new_dentry: destination 4061 * @delegated_inode: returns an inode needing a delegation break 4062 * @flags: rename flags 4063 * 4064 * The caller must hold multiple mutexes--see lock_rename()). 4065 * 4066 * If vfs_rename discovers a delegation in need of breaking at either 4067 * the source or destination, it will return -EWOULDBLOCK and return a 4068 * reference to the inode in delegated_inode. The caller should then 4069 * break the delegation and retry. Because breaking a delegation may 4070 * take a long time, the caller should drop all locks before doing 4071 * so. 4072 * 4073 * Alternatively, a caller may pass NULL for delegated_inode. This may 4074 * be appropriate for callers that expect the underlying filesystem not 4075 * to be NFS exported. 4076 * 4077 * The worst of all namespace operations - renaming directory. "Perverted" 4078 * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 4079 * Problems: 4080 * a) we can get into loop creation. 4081 * b) race potential - two innocent renames can create a loop together. 4082 * That's where 4.4 screws up. Current fix: serialization on 4083 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 4084 * story. 4085 * c) we have to lock _four_ objects - parents and victim (if it exists), 4086 * and source (if it is not a directory). 4087 * And that - after we got ->i_mutex on parents (until then we don't know 4088 * whether the target exists). Solution: try to be smart with locking 4089 * order for inodes. We rely on the fact that tree topology may change 4090 * only under ->s_vfs_rename_mutex _and_ that parent of the object we 4091 * move will be locked. Thus we can rank directories by the tree 4092 * (ancestors first) and rank all non-directories after them. 4093 * That works since everybody except rename does "lock parent, lookup, 4094 * lock child" and rename is under ->s_vfs_rename_mutex. 4095 * HOWEVER, it relies on the assumption that any object with ->lookup() 4096 * has no more than 1 dentry. If "hybrid" objects will ever appear, 4097 * we'd better make sure that there's no link(2) for them. 4098 * d) conversion from fhandle to dentry may come in the wrong moment - when 4099 * we are removing the target. Solution: we will have to grab ->i_mutex 4100 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4101 * ->i_mutex on parents, which works but leads to some truly excessive 4102 * locking]. 4103 */ 4104 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 4105 struct inode *new_dir, struct dentry *new_dentry, 4106 struct inode **delegated_inode, unsigned int flags) 4107 { 4108 int error; 4109 bool is_dir = d_is_dir(old_dentry); 4110 const unsigned char *old_name; 4111 struct inode *source = old_dentry->d_inode; 4112 struct inode *target = new_dentry->d_inode; 4113 bool new_is_dir = false; 4114 unsigned max_links = new_dir->i_sb->s_max_links; 4115 4116 if (source == target) 4117 return 0; 4118 4119 error = may_delete(old_dir, old_dentry, is_dir); 4120 if (error) 4121 return error; 4122 4123 if (!target) { 4124 error = may_create(new_dir, new_dentry); 4125 } else { 4126 new_is_dir = d_is_dir(new_dentry); 4127 4128 if (!(flags & RENAME_EXCHANGE)) 4129 error = may_delete(new_dir, new_dentry, is_dir); 4130 else 4131 error = may_delete(new_dir, new_dentry, new_is_dir); 4132 } 4133 if (error) 4134 return error; 4135 4136 if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 4137 return -EPERM; 4138 4139 if (flags && !old_dir->i_op->rename2) 4140 return -EINVAL; 4141 4142 /* 4143 * If we are going to change the parent - check write permissions, 4144 * we'll need to flip '..'. 4145 */ 4146 if (new_dir != old_dir) { 4147 if (is_dir) { 4148 error = inode_permission(source, MAY_WRITE); 4149 if (error) 4150 return error; 4151 } 4152 if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4153 error = inode_permission(target, MAY_WRITE); 4154 if (error) 4155 return error; 4156 } 4157 } 4158 4159 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 4160 flags); 4161 if (error) 4162 return error; 4163 4164 old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4165 dget(new_dentry); 4166 if (!is_dir || (flags & RENAME_EXCHANGE)) 4167 lock_two_nondirectories(source, target); 4168 else if (target) 4169 mutex_lock(&target->i_mutex); 4170 4171 error = -EBUSY; 4172 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4173 goto out; 4174 4175 if (max_links && new_dir != old_dir) { 4176 error = -EMLINK; 4177 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4178 goto out; 4179 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4180 old_dir->i_nlink >= max_links) 4181 goto out; 4182 } 4183 if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4184 shrink_dcache_parent(new_dentry); 4185 if (!is_dir) { 4186 error = try_break_deleg(source, delegated_inode); 4187 if (error) 4188 goto out; 4189 } 4190 if (target && !new_is_dir) { 4191 error = try_break_deleg(target, delegated_inode); 4192 if (error) 4193 goto out; 4194 } 4195 if (!old_dir->i_op->rename2) { 4196 error = old_dir->i_op->rename(old_dir, old_dentry, 4197 new_dir, new_dentry); 4198 } else { 4199 WARN_ON(old_dir->i_op->rename != NULL); 4200 error = old_dir->i_op->rename2(old_dir, old_dentry, 4201 new_dir, new_dentry, flags); 4202 } 4203 if (error) 4204 goto out; 4205 4206 if (!(flags & RENAME_EXCHANGE) && target) { 4207 if (is_dir) 4208 target->i_flags |= S_DEAD; 4209 dont_mount(new_dentry); 4210 detach_mounts(new_dentry); 4211 } 4212 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4213 if (!(flags & RENAME_EXCHANGE)) 4214 d_move(old_dentry, new_dentry); 4215 else 4216 d_exchange(old_dentry, new_dentry); 4217 } 4218 out: 4219 if (!is_dir || (flags & RENAME_EXCHANGE)) 4220 unlock_two_nondirectories(source, target); 4221 else if (target) 4222 mutex_unlock(&target->i_mutex); 4223 dput(new_dentry); 4224 if (!error) { 4225 fsnotify_move(old_dir, new_dir, old_name, is_dir, 4226 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4227 if (flags & RENAME_EXCHANGE) { 4228 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4229 new_is_dir, NULL, new_dentry); 4230 } 4231 } 4232 fsnotify_oldname_free(old_name); 4233 4234 return error; 4235 } 4236 EXPORT_SYMBOL(vfs_rename); 4237 4238 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4239 int, newdfd, const char __user *, newname, unsigned int, flags) 4240 { 4241 struct dentry *old_dir, *new_dir; 4242 struct dentry *old_dentry, *new_dentry; 4243 struct dentry *trap; 4244 struct nameidata oldnd, newnd; 4245 struct inode *delegated_inode = NULL; 4246 struct filename *from; 4247 struct filename *to; 4248 unsigned int lookup_flags = 0; 4249 bool should_retry = false; 4250 int error; 4251 4252 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4253 return -EINVAL; 4254 4255 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 4256 (flags & RENAME_EXCHANGE)) 4257 return -EINVAL; 4258 4259 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 4260 return -EPERM; 4261 4262 retry: 4263 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); 4264 if (IS_ERR(from)) { 4265 error = PTR_ERR(from); 4266 goto exit; 4267 } 4268 4269 to = user_path_parent(newdfd, newname, &newnd, lookup_flags); 4270 if (IS_ERR(to)) { 4271 error = PTR_ERR(to); 4272 goto exit1; 4273 } 4274 4275 error = -EXDEV; 4276 if (oldnd.path.mnt != newnd.path.mnt) 4277 goto exit2; 4278 4279 old_dir = oldnd.path.dentry; 4280 error = -EBUSY; 4281 if (oldnd.last_type != LAST_NORM) 4282 goto exit2; 4283 4284 new_dir = newnd.path.dentry; 4285 if (flags & RENAME_NOREPLACE) 4286 error = -EEXIST; 4287 if (newnd.last_type != LAST_NORM) 4288 goto exit2; 4289 4290 error = mnt_want_write(oldnd.path.mnt); 4291 if (error) 4292 goto exit2; 4293 4294 oldnd.flags &= ~LOOKUP_PARENT; 4295 newnd.flags &= ~LOOKUP_PARENT; 4296 if (!(flags & RENAME_EXCHANGE)) 4297 newnd.flags |= LOOKUP_RENAME_TARGET; 4298 4299 retry_deleg: 4300 trap = lock_rename(new_dir, old_dir); 4301 4302 old_dentry = lookup_hash(&oldnd); 4303 error = PTR_ERR(old_dentry); 4304 if (IS_ERR(old_dentry)) 4305 goto exit3; 4306 /* source must exist */ 4307 error = -ENOENT; 4308 if (d_is_negative(old_dentry)) 4309 goto exit4; 4310 new_dentry = lookup_hash(&newnd); 4311 error = PTR_ERR(new_dentry); 4312 if (IS_ERR(new_dentry)) 4313 goto exit4; 4314 error = -EEXIST; 4315 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 4316 goto exit5; 4317 if (flags & RENAME_EXCHANGE) { 4318 error = -ENOENT; 4319 if (d_is_negative(new_dentry)) 4320 goto exit5; 4321 4322 if (!d_is_dir(new_dentry)) { 4323 error = -ENOTDIR; 4324 if (newnd.last.name[newnd.last.len]) 4325 goto exit5; 4326 } 4327 } 4328 /* unless the source is a directory trailing slashes give -ENOTDIR */ 4329 if (!d_is_dir(old_dentry)) { 4330 error = -ENOTDIR; 4331 if (oldnd.last.name[oldnd.last.len]) 4332 goto exit5; 4333 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len]) 4334 goto exit5; 4335 } 4336 /* source should not be ancestor of target */ 4337 error = -EINVAL; 4338 if (old_dentry == trap) 4339 goto exit5; 4340 /* target should not be an ancestor of source */ 4341 if (!(flags & RENAME_EXCHANGE)) 4342 error = -ENOTEMPTY; 4343 if (new_dentry == trap) 4344 goto exit5; 4345 4346 error = security_path_rename(&oldnd.path, old_dentry, 4347 &newnd.path, new_dentry, flags); 4348 if (error) 4349 goto exit5; 4350 error = vfs_rename(old_dir->d_inode, old_dentry, 4351 new_dir->d_inode, new_dentry, 4352 &delegated_inode, flags); 4353 exit5: 4354 dput(new_dentry); 4355 exit4: 4356 dput(old_dentry); 4357 exit3: 4358 unlock_rename(new_dir, old_dir); 4359 if (delegated_inode) { 4360 error = break_deleg_wait(&delegated_inode); 4361 if (!error) 4362 goto retry_deleg; 4363 } 4364 mnt_drop_write(oldnd.path.mnt); 4365 exit2: 4366 if (retry_estale(error, lookup_flags)) 4367 should_retry = true; 4368 path_put(&newnd.path); 4369 putname(to); 4370 exit1: 4371 path_put(&oldnd.path); 4372 putname(from); 4373 if (should_retry) { 4374 should_retry = false; 4375 lookup_flags |= LOOKUP_REVAL; 4376 goto retry; 4377 } 4378 exit: 4379 return error; 4380 } 4381 4382 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4383 int, newdfd, const char __user *, newname) 4384 { 4385 return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4386 } 4387 4388 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 4389 { 4390 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 4391 } 4392 4393 int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4394 { 4395 int error = may_create(dir, dentry); 4396 if (error) 4397 return error; 4398 4399 if (!dir->i_op->mknod) 4400 return -EPERM; 4401 4402 return dir->i_op->mknod(dir, dentry, 4403 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4404 } 4405 EXPORT_SYMBOL(vfs_whiteout); 4406 4407 int readlink_copy(char __user *buffer, int buflen, const char *link) 4408 { 4409 int len = PTR_ERR(link); 4410 if (IS_ERR(link)) 4411 goto out; 4412 4413 len = strlen(link); 4414 if (len > (unsigned) buflen) 4415 len = buflen; 4416 if (copy_to_user(buffer, link, len)) 4417 len = -EFAULT; 4418 out: 4419 return len; 4420 } 4421 EXPORT_SYMBOL(readlink_copy); 4422 4423 /* 4424 * A helper for ->readlink(). This should be used *ONLY* for symlinks that 4425 * have ->follow_link() touching nd only in nd_set_link(). Using (or not 4426 * using) it for any given inode is up to filesystem. 4427 */ 4428 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4429 { 4430 struct nameidata nd; 4431 void *cookie; 4432 int res; 4433 4434 nd.depth = 0; 4435 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4436 if (IS_ERR(cookie)) 4437 return PTR_ERR(cookie); 4438 4439 res = readlink_copy(buffer, buflen, nd_get_link(&nd)); 4440 if (dentry->d_inode->i_op->put_link) 4441 dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4442 return res; 4443 } 4444 EXPORT_SYMBOL(generic_readlink); 4445 4446 /* get the link contents into pagecache */ 4447 static char *page_getlink(struct dentry * dentry, struct page **ppage) 4448 { 4449 char *kaddr; 4450 struct page *page; 4451 struct address_space *mapping = dentry->d_inode->i_mapping; 4452 page = read_mapping_page(mapping, 0, NULL); 4453 if (IS_ERR(page)) 4454 return (char*)page; 4455 *ppage = page; 4456 kaddr = kmap(page); 4457 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4458 return kaddr; 4459 } 4460 4461 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4462 { 4463 struct page *page = NULL; 4464 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 4465 if (page) { 4466 kunmap(page); 4467 page_cache_release(page); 4468 } 4469 return res; 4470 } 4471 EXPORT_SYMBOL(page_readlink); 4472 4473 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 4474 { 4475 struct page *page = NULL; 4476 nd_set_link(nd, page_getlink(dentry, &page)); 4477 return page; 4478 } 4479 EXPORT_SYMBOL(page_follow_link_light); 4480 4481 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 4482 { 4483 struct page *page = cookie; 4484 4485 if (page) { 4486 kunmap(page); 4487 page_cache_release(page); 4488 } 4489 } 4490 EXPORT_SYMBOL(page_put_link); 4491 4492 /* 4493 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 4494 */ 4495 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 4496 { 4497 struct address_space *mapping = inode->i_mapping; 4498 struct page *page; 4499 void *fsdata; 4500 int err; 4501 char *kaddr; 4502 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 4503 if (nofs) 4504 flags |= AOP_FLAG_NOFS; 4505 4506 retry: 4507 err = pagecache_write_begin(NULL, mapping, 0, len-1, 4508 flags, &page, &fsdata); 4509 if (err) 4510 goto fail; 4511 4512 kaddr = kmap_atomic(page); 4513 memcpy(kaddr, symname, len-1); 4514 kunmap_atomic(kaddr); 4515 4516 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4517 page, fsdata); 4518 if (err < 0) 4519 goto fail; 4520 if (err < len-1) 4521 goto retry; 4522 4523 mark_inode_dirty(inode); 4524 return 0; 4525 fail: 4526 return err; 4527 } 4528 EXPORT_SYMBOL(__page_symlink); 4529 4530 int page_symlink(struct inode *inode, const char *symname, int len) 4531 { 4532 return __page_symlink(inode, symname, len, 4533 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 4534 } 4535 EXPORT_SYMBOL(page_symlink); 4536 4537 const struct inode_operations page_symlink_inode_operations = { 4538 .readlink = generic_readlink, 4539 .follow_link = page_follow_link_light, 4540 .put_link = page_put_link, 4541 }; 4542 EXPORT_SYMBOL(page_symlink_inode_operations); 4543